Air Sentinel: An IoT-Based Platform for Monitoring Indoor Air Quality in Elementary Schools of the Global South

Air Sentinel: An IoT-Based Platform for Monitoring Indoor Air Quality in Elementary Schools of the Global South | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Air Sentinel: An IoT-Based Platform for Monitoring Indoor Air Quality in Elementary Schools of the Global South Salvador Ruiz-Correa, Rubén López-Revilla, Lina Riego-Ruiz, Gladys Díaz, and 8 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8379552/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 4 You are reading this latest preprint version Abstract Indoor air quality (IAQ) is neglected in schools of the Global South. Carbon dioxide (CO 2 ) and particulate matter (PM 2.5 ) are indicators of ventilation and exposure risk to airborne infections, respectively. We deployed Air Sentinel , a decentralized mobile-driven IoT network, to monitor these indicators across 106 spaces in 53 elementary schools in the San Luis Potosí Metropolitan Area, Mexico. Machine learning was used to forecast CO 2 and PM 2.5 concentrations one hour in advance, and the Wells-Riley model, based on CO 2 concentration, to estimate airborne infection probability (AIP). IAQ was poor, acceptable, and excellent in 27.3%, 36.4%, and 67.2% of the classrooms, respectively. During the cold season, 97.67% of the classroom CO 2 levels were typical (400–1000 ppm); in the hot season, 67.67% of the classroom CO 2 levels were typical, and 1.19% exceeded the high exposure threshold (> 2000 ppm). Classroom CO 2 dynamics exhibited low temporal synchrony. The strongest forecasting performance for CO 2 occurred in the hot season, but the PM 2.5 forecast failed in either season while AIP increased during the first two hours of class in both seasons. The successful CO 2 forecasting model has potential for real-time IAQ management in the cold season. The failure to forecast PM 2.5 levels suggests that localized sources drive their dynamics. We conclude that the Air Sentinel network is a convenient classroom IAQ monitor in the Global South. Earth and environmental sciences/Climate sciences Earth and environmental sciences/Environmental sciences Earth and environmental sciences/Environmental social sciences Indoor air quality Carbon dioxide forecasting Classroom ventilation PM2.5 exposure Machine learning regression Airborne infection probability IoT environmental monitoring Figures Figure 1 Figure 2 Figure 3 Introduction In recent years, numerous studies have shown that indoor air can pose greater health risks than outdoor air. 1 In elementary schools, indoor air quality (IAQ) is a critical yet often overlooked factor influencing children’s respiratory health. 2 – 4 Classrooms frequently harbor a range of indoor air pollutants, including carbon dioxide (CO₂), fine particulate matter (PM 2.5 ), volatile organic compounds, and allergens, all of which can impact children's health and learning. International guidelines and standards, such as those of the World Health Organization and the American Society of Heating, Refrigerating, and Air-Conditioning Engineers, define thresholds for indoor pollutants that many schools fail to meet. 5 – 7 CO 2 and PM 2.5 are key indoor pollutants closely tied to ventilation quality and respiratory health. 2 , 3 Elevated CO₂ levels, mainly from human respiration in crowded or poorly ventilated spaces, signal inadequate air exchange and are linked to reduced student cognitive performance 8 and increased airborne infection risk. 9 Poor ventilation allows pathogens like influenza and SARS-CoV-2 to accumulate, while effective natural or mechanical systems improve IAQ by diluting contaminants. 10 In classrooms, proper airflow not only reduces the transmission of aerosols but also enhances thermal comfort. 8 , 11 , 12 PM 2.5 , originating from outdoor sources, cleaning, and resuspension, also poses serious health risks, especially for children with developing lungs and immune systems. 2 , 13 Exposure is linked to asthma, bronchitis, impaired lung function, and long-term cardiovascular issues. Together, CO₂ and PM 2.5 are vital IAQ indicators, and their systematic monitoring supports targeted ventilation strategies and broader interventions for healthier classroom environments. Studies in the Global South have employed continuous monitoring and cross-sectional sampling to assess school IAQ. In Colombia, PM 2.5 data paired with activity logs and respiratory surveys showed that younger children inhale higher pollutant doses, often exceeding safe limits and reporting more symptoms. 14 Chilean schools with natural ventilation recorded CO₂ levels regularly above 1400 ppm, peaking at 5000 ppm in winter due to seasonal, architectural, and occupancy factors. 15 , 16 In Turkey, classrooms exhibited median CO₂ levels above 1000 ppm and PM 2.5 averages of 31.8 µg/m³, which correlated with increased respiratory symptoms and an increased risk of bronchitis. 17 Research from Portugal and Spain found that classroom activities, such as cleaning and window use, can cause sharp, short-term spikes in PM2.5. 18 Despite growing global attention to IAQ, Mexican cities, particularly the San Luis Potosí Metropolitan Area (SLPMA), remain underrepresented in research and policy. Situated in a dusty, semi-arid region, 19 SLPMA’s elementary schools face compounded IAQ risks from rapid urbanization, industrial emissions (Fig. 1 ), and deteriorating infrastructure. These conditions threaten children's respiratory health and academic outcomes; however, the absence of systematic monitoring hampers the identification of high-risk classrooms and the development of targeted interventions. Expanding IAQ research in cities like SLPMA is crucial for safeguarding vulnerable populations and guiding context-specific solutions. While active and passive sampling methods have been used for over two decades, the rise of low-cost online sensors offers new, affordable avenues for IAQ monitoring, 20–22 which is a vital development for the Global South, where financial constraints often limit access to conventional monitoring systems. This study evaluated ventilation and IAQ across 106 spaces, including classrooms and other rooms, within 53 of the 315 public elementary schools in the SLPMA. 23 By analyzing CO₂ and PM 2.5 concentrations, the research aimed to identify significant risk factors and offer evidence-based recommendations to improve health and safety in learning environments. A set of targeted research questions guided the study: How adequate is classroom ventilation in public elementary schools in San Luis Potosí? How do CO₂ and PM 2.5 levels fluctuate across seasons in classrooms? How do seasonal dynamics in indoor CO₂ and PM 2.5 concentrations across classrooms shape the reliability and generalizability of data-driven forecasting machine learning models? To address these questions, this study presents a scalable model for low-cost IAQ monitoring in resource-limited settings, using the SLPMA as a case example. By addressing seasonal patterns, pollutant forecasting, and ventilation strategies, it responds to structural and climatic challenges common across emergent economies in the Global South. The integration of real-time data and predictive modeling enables local stakeholders to enhance children's health and learning conditions, thereby promoting environmental justice and public health equity. Results Sampled Schools Our study sample comprises 53 elementary public schools selected by the local education ministry, with a focus on those located in areas with high COVID-19 incidence during the pandemic. Figure 2 shows population density across the SLPMA, with colored stars marking sampled schools and polygons representing statistical geographic units (AGEBs) defined by Instituto Nacional de Estadística y Geografía (INEGI). 23 Among these schools, 14.0% are located in very high-density areas, 18.7% in high-density areas, 33.3% in medium-density areas, 27.0% in low-density areas, and 6.4% in very low-density areas. Figure S1 displays the Marginalization Index (MI) across the SLPMA, developed by the Consejo Nacional de Población to measure social and economic deprivation using indicators in education, housing, income, and population distribution. 23 Colored stars mark sampled schools, with 50.0% in very low marginalization areas, 37.5% in low, and 12.5% in medium marginalization zones. Air Sentinel Deployment Air Sentinel was installed with parental consent in 11 schools to continuously monitor CO₂, PM 2.5 , temperature, and humidity (Figure S2). One classroom was selected per school, and data were collected during four epidemiological weeks in 2023: weeks 23 and 25 (hot season, 13–30 o C), 46 and 48 (cold season, 5–14 o C). Although limited connectivity affected data quality in weeks 25 and 46, CO₂ and PM 2.5 were successfully captured across all four weeks. Primary analyses use complete datasets from weeks 23 and 48, while CO₂ and PM 2.5 forecasting incorporates. Geometric Features and Ventilation Metrics of Classrooms The study assessed 73 classrooms and 33 additional rooms (e.g., libraries, labs) across 53 schools, using nine descriptors. Table S1 describes the geometric features of the 106 spaces: most had rectangular bases (only one was pentagonal). In all spaces, the window area exceeded 10% of the floor area, and windows were evenly split between sliding and top-hinged types, meeting lighting and ventilation standards. Maintenance issues affected 56% of the spaces; 39.2% lacked cross-ventilation. Occupancy density averaged 1.13 ± 0.33 per m² in classrooms and 1.07 ± 0.37 in other rooms, both of which are above the legal minimum of 0.625. Table 1 presents ventilation data for the 106 spaces. Forced ventilation was more common in classrooms (68.5%) than in other rooms (31.5%). IAQ ratings showed disparities: poor in 27.3% of the classrooms compared to 1.3% in other rooms, and excellent in 67.2% compared to 34.6% in other rooms. Acceptable IAQ ratings were similar for classrooms (31.5%) and other rooms (36.4%). Figure S3 shows the variability in ventilation scores (box plots). The magnitude of scores and IAQ levels of all school rooms is mapped in Figures S4 and S5 using scaled, shaded circles. CO and PM Concentration Metrics CO₂ and PM 2.5 were continuously monitored in 11 schools during class hours (08:00–14:00 h) using Air Sentinel . Table S2 summarizes the seasonal variation in CO₂ and PM 2.5 concentrations across monitored school spaces (Figure S6). CO₂ levels were higher during the cold season, with a mean of 894.45 ppm (CV = 0.44), compared to 666.67 ppm (CV = 0.21) in the hot season. The cold season also exhibited greater dispersion, with a maximum value of 3069.85 ppm, more than double that of the hot season (1226.38 ppm), and a wider interquartile range (588.44–1105.44 ppm). PM 2.5 concentrations showed an inverse seasonal pattern. Mean levels were higher in the hot season (14.66 µg/m³) than in the cold season (9.67 µg/m³), and variability was greater in the latter (CV = 1.30 vs. 0.70). The cold season also presented extreme outliers, with a maximum of 853.26 µg/m³, far exceeding typical indoor thresholds and the hot season peak of 58.76 µg/m³. These findings suggest distinct seasonal dynamics in IAQ, with elevated CO₂ during colder months linked to reduced ventilation, and sporadic PM 2.5 spikes in the cold season, potentially driven by localized sources or infiltration events. Pairwise Pearson’s coefficient (PC) for CO₂ time series across classrooms averaged 0.24 ± 0.25, with a mean R² = 0.12, indicating low linear synchrony and predictive power across classrooms. Cold season values were slightly higher (PC = 0.26 ± 0.24; R² = 0.13), but still modest. PM 2.5 analysis showed even weaker linear associations. In the hot season, PC averaged 0.10 ± 0.42, and R² was 0.19; in the cold season; PC dropped to 0.06 ± 0.29, and R² decreased to 0.08, confirming limited linear temporal associations and predictability across classrooms. Forecasting CO and PM Concentrations During the hot season, an XGBoost autoregressive model 26 achieved optimal CO₂ forecasting with an 8-hour lookback window ( h = 8). Using leave-one-classroom-out cross-validation, it achieved an average R² of 0.85, MAE of 12.12 ppm, and RMSE of 17.44 ppm. Generalization gaps were 10.76 ppm (MAE) and 15.63 ppm (RMSE), indicating low overfitting but acceptable robustness for deployment in unseen classrooms. In the cold season, the XGBoost model optimal forecasting setup showed slightly reduced performance. Using leave-one-classroom-out cross-validation, the model achieved an average R² of 0.80, with an MAE of 68.04 ppm and an RMSE of 95.7 ppm, indicating larger prediction deviations. The generalization gaps were 62.95 ppm (MAE) and 88.23 ppm (RMSE), suggesting lower generalizability under colder conditions. Applying the XGBoost forecasting framework to PM 2.5 concentrations yielded weak results across both hot and cold seasons, with low R² values and high prediction errors, indicating that short-term PM 2.5 variability cannot be reliably predicted using historical indoor data alone. These findings suggest that PM 2.5 dynamics in schools are driven by external factors such as traffic emissions, cleaning, or resuspension events, which are not captured by the model’s input features. CO Concentration Trends Beyond aggregate analyses, a detailed assessment of pollutant trends was conducted at the individual classroom level across the 11 schools. Metrics such as the coefficient of variation (CV) and Pearson’s coefficient (PC) between CO₂ and PM 2.5 were used to capture intra-room variability and identify classrooms with unstable or elevated profiles. Due to the volume and granularity of the data, classroom-specific statistics are provided in Tables S3–S6. This deeper analysis enhances understanding of IAQ dynamics and supports targeted interventions for specific classroom conditions. CO 2 Concentration Metrics and Infection Probability Figure 4 compares average CO₂-based AIP during class hours across schools in the hot (top panel) and cold (bottom panel) seasons. In both cases, AIP values rose exponentially, reaching ~ 0.9 after 150 minutes in the hot season and ~ 0.95 in the cold season, due to higher CO₂ levels from reduced ventilation. These values were similar across classrooms. Each curve was modeled using \(\:y=A(1-{e}^{-Bx})\) (Table S7). A representative case study is illustrated in Figure S7. XGBoost forecasting showed high predictive accuracy for short-term infection probability during the hot season, using a leave-one-classroom-out strategy to ensure generalizability. It achieved a test R² of 0.98, with an MAE of 0.005 and an RMSE of 0.008. The generalization gap was minimal (MAE = 0.004; RMSE = 0.006), indicating strong transferability across diverse classroom conditions. During the cold season, the model maintained robust performance with tighter predictive alignment. Using the leave-one-classroom-out approach, it achieved a test R² of 0.99, indicating near-perfect correlation between predicted and observed AIP values. Test MAE and RMSE remained at 0.005 and 0.008, with a low generalization gap (MAE = 0.004; RMSE = 0.006). These results confirm the model’s robustness under seasonal IAQ shifts, particularly elevated CO₂ levels resulting from reduced ventilation, without compromising predictive reliability. Outdoor PM Pollution Sources Although PM 2.5 concentrations did not reveal consistent statistical patterns across classrooms, the absence of spatial coherence suggests the need to investigate localized pollution and emission sources (for which quantitative data is available). In a complex urban setting like the SLPMA, local PM 2.5 levels are significantly influenced by the surrounding environment. A detailed spatial inventory reveals a diverse array of potential contributors distributed across the city: 143 gas stations, 129 brick kilns, 123 biomass burning sites, 90 carpentry shops, 62 solvent shops, 50 chemical factories, 25 mining operations, 15 incinerators, seven paper factories, seven pesticide factories, six cement factories, six landfills and a concentration of additional emission sources in the southern region. These sources exhibit varying distributions, contributing to air quality heterogeneity (Fig. 1 ). A buffer zone analysis was performed using radii ranging from 100 meters to 3 km in 100-meter increments, generating 30 concentric zones around each school to assess the density of nearby pollution sources (Figure S8). For each buffer, pollution sources were quantified using the georeferenced inventory. Additionally, street segments within 100 meters of each school were counted to estimate the potential traffic-related exposure (Figure S9). Sources from nearby streets may emit NO₂, CO, black carbon, and PM 2.5 , especially in congested areas. This spatial analysis serves as a proxy for exposure risk, helping to identify environmental pressure zones, and guide future monitoring studies and mitigation strategies. Table S4 provides a detailed summary of stationary pollution sources located within a 3 km radius of each elementary school in the SPLMA. Table S5 complements this finding by listing all inventoried emission sources, both stationary and mobile, within the same spatial extent, offering a comprehensive view of the surrounding environmental pressure at the broadest buffer scale considered. A buffer size of 3 km offers a broad spatial context, encompassing not only immediate surroundings but also more distant contributors. Discussion Elementary schools in the SLPMA are predominantly located in medium- and low-density zones (> 60%), indicating concentration in suburban or peri-urban areas. Nearly one-third are in high- or very high-density zones, where environmental stressors such as air pollution and traffic congestion are more intense, while only 6.4% are in very low-density areas, suggesting gaps in semirural access. This pattern aligns with international findings; Sadrizadeh et al. 35 reviewed 304 studies (1970–2022) showing that IAQ disparities often align with urbanization levels, since urban schools exhibiting elevated CO₂, particulate matter, and VOCs levels due to traffic, poor ventilation, and aging infrastructure. Window surface areas exceeding 10% of the classroom floor areas meet foundational design standards for natural lighting and ventilation, consistent with ASHRAE and International Residential Code recommendations. However, this does not ensure adequate ventilation, as actual performance depends on the operability and strategic placement of the openings. 36 While the presence of both sliding and top-hinged windows allows for varied ventilation strategies, 56% of spaces suffer from poor maintenance, compromising window operability and IAQ. Neglected mechanical components can hinder airflow, trap pollutants, and compromise thermal comfort, 37 and in schools where ventilation is vital for cognitive function and infection control, non-functional windows represent lost opportunities for passive environmental regulation. Insufficient cross-ventilation affects 39.2% of the classrooms, posing a significant barrier to adequate natural ventilation. Cross-ventilation through openings on opposing walls is crucial for enhancing air exchange and reducing pollutants, leading to lower CO₂ levels and better thermal comfort than single-sided ventilation. 38 In San Luis Potosí schools, where mechanical systems may be inconsistently available, architectural solutions are critical. Reported occupancy densities of 1.13 ± 0.33 students/m² in classrooms and 1.07 ± 0.37 persons/m² in other rooms comply with the Mexican legal standard of 0.625 students/m². However, regulatory compliance does not guarantee optimal conditions, as even moderate densities can elevate CO₂ levels if ventilation is inadequate. 39 Forced ventilation systems are unevenly distributed; present in 68.5% of classrooms but only in 48.5% of other rooms, revealing infrastructure disparities that may reflect focus on instructional areas over administrative ones. 40 Limited coverage in non-classroom areas suggests planning blind spots that could impact staff health and building performance. Our results revealed significant seasonal variability in IAQ, particularly in CO₂ and PM 2.5 concentrations, closely tied to environmental conditions and ventilation dynamics during class periods (08:00–14:00 hours). CO₂ concentrations were higher during the cold season (mean = 894.45 ppm) compared to the hot season (666.67 ppm) due to reduced natural ventilation from closed windows and doors, with greater variability (CV = 0.44 vs. 0.21). These results align with elevated winter CO₂ levels in naturally ventilated classrooms 2 , 41 Low pairwise Pearson coefficients (PC = 0.24–0.26) and low R² values (0.12–0.13) indicated weak linear temporal synchrony between classrooms, suggesting exploration of non-linear relations using machine learning tools. PM 2.5 concentrations varied seasonally, with higher mean levels in the hot season (14.66 µg/m³) than the cold season (9.67 µg/m³) due to increased outdoor air exchange. However, the cold season exhibited significantly higher maximum values (853.26 µg/m³) and greater variability (CV = 1.30 vs. 0.70), suggesting episodic spikes from localized sources. This aligns with prior research showing higher summer PM 2.5 from outdoor infiltration while winter peaks were attributed to indoor sources, with chalk, cleaning, socioeconomic factors, and combustion identified as contributors. 44 During the hot season, XGBoost achieved strong CO₂ predictive performance with an 8-hour lookback window (R² = 0.85; MAE = 12.12 ppm; RMSE = 17.44 ppm), with moderate generalization gaps suggesting limited overfitting and good robustness. 45 , 46 During the cold season, performance declined (R² = 0.80; MAE = 68.04 ppm; RMSE = 95.7 ppm) with generalization gaps indicating reduced adaptability, likely due to variable ventilation behaviors introducing non-linearities. 47 , 48 PM 2.5 forecasting yielded poor results in both seasons, with low R² values and high errors, suggesting that levels are driven by complex, short-term fluctuations not captured by historical indoor data. External factors such as traffic emissions, cleaning, and particle resuspension play significant roles, 49,50 emphasizing the need to map surrounding pollution sources. PM 2.5 data from 11 schools during the spring and winter of 2023 revealed marked seasonal variability. 51 In the cold season, several classrooms exceeded WHO's 24-hour threshold of 15 µg/m³, 52 with schools HC, FS, and FN exhibiting repeated high concentrations (HC: 53.1 µg/m³; FS: 34.2 µg/m³) likely due to limited ventilation and proximity to pollution sources. Extreme outliers (JMJ: 853.3 µg/m³; VFS: 456.9 µg/m³) underscore the need for continuous monitoring. 53 In the hot season, most classrooms maintained levels below 20 µg/m³, though hotspots like JMJ, HC, and IZ exceeded 25 µg/m³ (peak: 58.4 µg/m³), suggesting persistent challenges. 54 Forecasting analysis indicates that CO₂ concentrations can be predicted with moderate accuracy across both seasons, opening the door to CO₂-based modeling of airborne infection risk for respiratory illnesses, as CO₂ reflects occupancy and air exchange. In contrast, PM 2.5 forecasts performed poorly in both seasons, with variability stemming from localized, episodic factors not captured by indoor historical data, emphasizing the need to map external pollution sources. CO₂-based infection probability analysis reveals clear seasonal patterns with significant implications for IAQ and infection risk management. Infection probability during class hours follows an exponential growth curve in both seasons, with the cold season showing a steeper trajectory due to higher CO₂ levels from reduced ventilation. The exponential saturation model effectively captures these dynamics, and the time to reach 95% infection probability is consistently shorter in the cold season. 55 , 56 XGBoost modeling achieved high accuracy in both the hot season (R² = 0.98; MAE = 0.005; RMSE = 0.008) and cold season (R² = 0.99), with strong generalizability confirmed by leave-one-classroom-out strategy, demonstrating adaptability to seasonal IAQ shifts and affirming the feasibility of using CO₂ data for real-time infection risk forecasting. Spatial buffer analysis using a 3 km radius effectively identifies environmental pressure zones and supports monitoring and policy planning. 57 – 60 Results reveal an average of 110.1 pollution sources per sampling point, with gas stations, biomass combustion sites, and carpentry shops most common. Brick kilns show extreme variability (mean = 9.5; max = 89), indicating industrial clusters with high particulate emissions. 61 High-impact but less frequent sources such as incinerators, cement factories, and landfills may exert disproportionate influence, and wide variability reflects spatial inequality in exposure. 58 CO₂ is widespread (138 sources) and often co-occurs with hazardous pollutants such as cadmium, mercury, lead, arsenic, and cyanide, 52,62 with lead standing out (mean = 5.0; max = 21). Even low concentrations of dioxins, furans, and VOCs raise concerns about long-term carcinogenic exposure. 63 Schools are closely surrounded by traffic infrastructure (average 8.4 streets within 100 meters), increasing exposure to PM 2.5 , NO₂, and O₃. Elevated NO₂ levels worsen asthma and impair lung function, 59,60 while ground-level ozone threatens respiratory health and contributes to chronic conditions. 64 The Air Sentinel platform successfully demonstrated the viability of a decentralized, mobile-driven IoT network for IAQ monitoring in resource-limited schools, enabling real-time data collection and visualization across classrooms using Bluetooth, Wi-Fi, and 4G connectivity. However, the limitations included Bluetooth's short range and interference, inconsistent Wi-Fi coverage, cellular signal variability and cost concerns, highlighting the need for more resilient, low-power communication solutions. Despite these constraints, Air Sentinel delivered valuable seasonal IAQ insights and actionable feedback for educators, affirming its scalability for school-based environmental health surveillance. Methods Location The SLPMA, comprising San Luis Potosí, Soledad de Graciano Sánchez, and nearby areas, is the state’s most populous and economically vital region. In 2020, its 1.2 million residents made it Mexico’s 11th largest metropolitan area, housing nearly one-third of the state’s population. As the leading industrial, commercial, and cultural center, it features strong manufacturing, growing coordination, and a strategic location in Mexico. Public elementary schools in this area are particularly vulnerable to IAQ issues due to the semi-arid climate, rapid urban growth, industrial emissions (Fig. 1 ), and deteriorating infrastructure with inadequate maintenance. 19 Ventilation assessments were carried out in 106 spaces across 53 schools in the SLPMA. IAQ measurements were conducted in one classroom per school within a subset of 11 schools, where parents provided signed consent forms authorizing the installation of measurement devices during class hours. Ventilation Metrics We applied the Harvard Healthy Buildings Program protocol 24 to estimate ventilation schools using the LinkApp mobile tool developed by our research team. Based on CO₂ decay, this method is suitable for naturally or mechanically ventilated areas where direct airflow measurement is challenging. An empty classroom is selected to avoid human-generated CO₂. Room characteristics, including use, size, occupancy, and openings, determine the number of calibrated monitors, which are placed 1–1.5 m above the ground and away from ventilation sources. Dry ice elevates CO₂, which is dispersed with fans until it reaches ~ 2000 ppm. Calibrated monitors (range 0–5000 ppm, accuracy ± 50 ppm) record CO₂ concentration every minute. The rate of decline indicates outdoor air exchange, and each sensor’s decay curve yields its air changes per hour (ACH). 24 The formula assumes a well-mixed indoor environment and constant outdoor CO 2 concentration, and uses the exponential decay model, $$\:ACH\:=\:-\frac{1}{\varDelta\:T}\text{log}\left(\frac{{C}_{f}-{C}_{a}}{{C}_{i}-{C}_{a}}\right),$$ where ACH is expressed in h − 1 , C f is the final CO 2 concentration in ppm, C i is the maximum CO 2 concentration in ppm, C a is the outdoor CO 2 concentration in ppm, and ∆ T is the time in minutes elapsed at which the decrease in CO 2 concentration is approximately 37% the maximum initial concentration. The ventilation rate is defined as: $$\:Ventilation\:rate\:=\:\frac{\:{ACH}_{min}\times\:\:V\:\times\:1000\:}{3600}\:\left(\frac{L}{s}\right),$$ where V is the room's volume and ACH min is the lowest ACH measurement obtained from the room monitors. The ventilation score is calculated as the ratio of ACH min to the required ventilation rate ( ACH req ), which represents the minimum airflow needed to remove the CO₂ generated by the occupants: $$\:score\:=\:\frac{{ACH}_{min\:}\:}{{ACH}_{req}}\:=\:\frac{{CFM}_{min\:}\:}{{CFM}_{req}}$$ A score ≥ 2 leads to an excellent IAQ, a score from 1 to 2 leads to an acceptable IAQ, and a score < 1 leads to a poor IAQ. 25 We computed descriptive statistics for ACH min , ACH max , ACH average , and the score in Table 2. CO 2 and PM 2.5 Concentration Metrics 2 We computed seasonal descriptive statistics, including the coefficient of variation (CV) for CO₂, PM 2.5 , temperature, and humidity across 11 classrooms. We assessed how much variability in one classroom’s CO₂ and PM 2.5 levels could be explained by the levels of another classroom across schools by a linear model. Mean excess CO₂ during class hours was calculated as: $$\:\varDelta\:C\:=\:\frac{1}{T}\:{\int\:}_{0}^{T}C\left(t\right)\:dt,$$ in ppm/min, where \(\:C\left(t\right)\) denotes the CO₂ instantaneous concentration above 1000 ppm. CO₂ levels of 400–1000 ppm was considered typical, 1000–2000 ppm moderate, and > 2000 ppm high.⁴ CO 2 and PM 2.5 Forecasting in Hot and Cold Seasons 2.5 We applied a machine learning approach to forecast indoor CO₂ and PM 2.5 concentrations one hour into the future, using the Extreme Gradient Boosting (XGBoost) regression. XGBoost is a scalable, regularized gradient boosting framework known for its high predictive accuracy and robustness to overfitting. 26 , 27 The input dataset comprised CO₂ (PM 2.5 ) measurements from 11 naturally ventilated classrooms, each monitored continuously during four nonconsecutive one-week periods, two periods representing the hot season (13°C-30°C) and two periods representing the cold season (5°C-14°C). This seasonal sampling design was intended to capture variability in ventilation behavior and occupancy patterns under contrasting climatic conditions. CO₂ (PM 2.5 ) concentrations were initially recorded at 1-minute intervals and subsequently aggregated into 5-minute averages to reduce high-frequency noise and align with operational decision-making times relevant to school environments. The forecasting task involved predicting CO₂ (PM 2.5 ) levels 12-time steps ahead (i.e., 60 minutes) using historical data from the preceding h hours. We evaluated multiple h configurations, ranging from 1 to 16. To ensure generalizability, model performance was assessed using a leave-one-classroom-out cross-validation strategy. For each iteration, the model was trained on data from 10 classrooms (two nonconsecutive weeks' data series per classroom) and tested on the excluded one. This process was repeated until each classroom's data had served as the test set. All reported metrics, including the coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE), reflect out-of-sample predictions on held-out classrooms. Hyperparameters were optimized using grid search. In addition to XGBoost, we tested several alternative regression algorithms, such as support vector regression and random forest. While some models performed adequately under specific configurations, XGBoost consistently outperformed them in terms of predictive accuracy and generalization across classrooms. A transformers-based approach was not considered in our study due to the limited size of our data set, for which XGBoost remains the more practical choice. This modeling strategy aligns with best practices in time series forecasting and environmental data science, where temporal generalization must be rigorously validated. 28 , 29 CO 2 Concentration Metrics and Airborne Infection Probability CO₂-based estimates of airborne infection probability (AIP) provide a practical proxy for ventilation adequacy and airborne transmission risk, particularly when direct airflow data are unavailable. In enclosed classrooms with limited mechanical ventilation, elevated CO₂ generated by exhaled aerosols indicates rebreathed air, potentially containing infectious particles. This method is particularly beneficial in colder seasons, when ventilation is reduced. Rudnick and Milton³⁰ adapted the Wells-Riley equation to use indoor CO₂ as a proxy for rebreathed air. The model estimates the rebreathed fraction from indoor–outdoor CO₂ differences and computes infection probability as: $$\:P\:=\:1\:-\:exp\left(-\frac{\stackrel{-}{f}Iq{\Delta\:}t}{n}\right),$$ where \(\:\stackrel{\prime }{f}\) is the average rebreathed fraction, \(\:I\:\) the number of infectors, \(\:q\:\) the quanta generation rate, \(\:{\Delta\:}t\) the exposure time, and \(\:n\) the number of occupants. See equations 3 and 9 in Rudnick and Milton. 30 We applied this model to real-time CO₂ data from 11 classrooms, assuming that \(\:I=1\) , \(\:q=1\:\) and \(\:n\:\) equal to student count. A second scenario included a 37% vaccination rate per school, which is the average rate in SLPMA. Bazant and Bush 31 offer an alternative to compute AIP. We also used XGBoost to forecast five minutes ahead using the prior ten minutes of data. This autoregressive model captured temporal risk patterns. To ensure generalizability, we employed leave-one-classroom-out cross-validation. Analysis focused on the first 120 minutes of class (08:00–10:00 h), when infection risk rose notably in both seasons. Outdoor PM 2.5 Pollution Sources Given the locations of pollution sources in the SLPMA, we conducted a spatial proximity analysis. We determined the distance between each classroom and nearby pollution sources to estimate potential exposure intensity, considering the proximity of a school to traffic corridors, industrial zones, or biomass burning sites. Even without emission rates, proximity itself is a strong indicator of influence, especially for PM 2.5 , which can vary sharply over short distances. We defined buffer zones around each school, ranging from 100 meters to 3 kilometers, to quantify the number of pollution sources within varying spatial scales. This approach enables the exploration of how local emission pressure may influence IAQ, even in the absence of direct emission measurements. By examining source density across these zones, we can construct a spatial narrative linking environmental context to the observed PM 2.5 data. Air Sentinel Deployment and Data Visualization A major challenge of our study was implementing an operational IoT network within the SLPMA, where connectivity and networking resources are limited. To address such limited infrastructure, the system was designed as a decentralized, mobile-driven network utilizing IAQ monitors that include calibrated sensors for CO 2 , PM 2.5 , humidity, and temperature. We paired each monitor with a cellphone via Bluetooth. Due to institutional regulations, each classroom was equipped with one monitor and only one cell phone, for which the teacher was responsible. Monitors were positioned strategically to measure breathing-zone air quality (1.0 and 1.5 meters above the floor), avoiding direct interference from pollution sources or factors that could distort measurements. In most cases, monitors were placed on the back wall or near the front wall, close to the teacher’s desk. 32 The monitor transmits real-time data to the companion mobile app ( LinkApp ), which stores, displays, and synchronizes data to the Air Sentinel cloud platform. The LinkApp manages data from up to 10 monitors. This lean and highly portable network structure with minimal hardware dependencies may be affected due to the short-range nature of Bluetooth and the variability of Wi-Fi signal strength. To support long-term analysis and remote data access, LinkApp was integrated into the cloud-based Air Sentinel backend platform that handled secure data ingestion through encrypted protocols and stored sensor readings in a structured database indexed by classroom and timestamp. It also provided alert management, historical trend analysis, and school-wide comparisons across seasons. The backend was designed to be scalable and compliant with data privacy standards. On the front end, LinkApp served as a sensor interface and teacher dashboard, offering immediate feedback on classroom IAQ with visual indicators reflecting safety thresholds (alerts at 1000 ppm and 35 \(\:\mu\:\) g/m ³ ). Additionally, a web-based Air Sentinel dashboard allowed school officials and researchers to access graphical summaries, download reports, and analyze exposure patterns over time. This dual-interface design empowered educators to make ventilation decisions during class while also enabling institutional oversight and public health integration. Educators accessed the Air Sentinel platform through a dedicated dashboard in two progressively enhanced versions; the initial version provided real-time visualization of raw CO₂ and PM 2.5 measurements, along with alert notifications triggered when predefined safety thresholds were exceeded. The second Air Sentinel iteration introduced a more advanced dashboard to analyze processed data and support decision-making. This dashboard allows users to filter by time, window, and classroom, tailoring the analysis to specific contexts. It presents average values for all monitored parameters (CO₂, PM 2.5 , humidity, and temperature) across the selected time and classifies each parameter qualitatively into categories. A dedicated section focuses on CO₂, featuring bar charts that display average CO₂ levels per classroom, enabling comparative analysis and identifying areas that may require improved ventilation. A horizontal summary chart categorizes all measurements into predefined CO₂ quality levels, offering a clear overview of the IAQ situation. A summary panel interprets the data by providing a numerical IAQ quality score and recommendations, such as moderate ventilation, to improve conditions. An explanatory legend connects CO₂ concentration ranges to health and comfort implications, helping stakeholders understand the significance of the observed values. Air Sentinel supports proactive IAQ management by adapting to diverse environments and data infrastructures, promoting healthier and safer indoor spaces through informed and responsive action. CO 2 , PM 2.5 Sensor Calibration and Measurements To ensure the reliability of IAQ measurements, we calibrated both CO 2 and PM 2.5 sensors before their deployment. For CO 2 sensors (range 0–5000 ppm, accuracy ± 50 ppm), which utilize non-dispersive infrared technology, calibration was conducted using a two-point reference procedure. Each device was exposed to a zero-gas environment (high-purity nitrogen) to establish the baseline; this was followed by exposure to a certified gas mixture with a known CO 2 concentration (typically 1000 ppm), allowing for span adjustment. This procedure established a calibration curve for each sensor, which was then validated against ambient outdoor air concentrations assumed to be near 400 ppm. In classrooms where automatic baseline correction algorithms were active, we reviewed sensor placement to ensure periodic exposure to fresh air, avoiding artificially elevated baselines due to continuous occupancy. 33 For PM 2.5 sensors (range 0-1000 \(\:\mu\:\) g/m ³ , accuracy ± 10% of the reading), calibration was conducted via co-location with a reference-grade gravimetric instrument in a controlled indoor setting over a period of two weeks. Sensors recorded optical particles count in parallel with mass concentrations obtained from the reference device. These datasets were analyzed to generate a correction factor using regression modeling, which accounted for temperature and relative humidity, as these factors influence particle scattering and sensor accuracy. Calibration models were embedded in the device firmware before field deployment. Additionally, we performed seasonal calibration validation to account for variations in aerosol composition and ambient moisture, characteristic of the semi-arid conditions in San Luis Potosí. 34 Temperature and relative humidity sensors were also calibrated using standard techniques. CO 2 , PM 2.5 , temperature, and relative humidity measurements were continuously recorded over 24-hour periods using monitors equipped with four dedicated sensors corresponding to each environmental parameter. Although data collection spanned the entire day, classroom activities occurred between 08:00 and 14:00 hours, Monday through Friday. Each monitor was connected via Bluetooth or Wi-Fi to the Air Sentinel platform, which enabled real-time data synchronization, centralized access, and secure storage. Data collection followed institutional standards for privacy protection. Experiments were conducted under the agreement of school authorities and parental consent. Declarations Competing interests The authors have no competing interests to declare. Author Contribution Conceptualization: SR‑C**,** RL‑R and LR‑R. Methodology: SR‑C and RL‑R. Software: SR‑C**,** RL‑R**,** LA**,** CH‑R**,** MAC‑J and JGR‑A. Validation: SR‑C**,** RL‑R**,** LR‑R and CH‑R. Formal analysis: SR‑C**,** RL‑R**,** LR‑R**,** GD**,** KB**,** NG‑H**,** AD**,** LA**,** FM‑C**,** CH‑R and MAC‑J. Investigation: SR‑C**,** RL‑R**,** LR‑R**,** GD**,** KB**,** FM‑C**,** CH‑R and MAC‑J. Resources: SR‑C**,** RL‑R**,** LR‑R**,** NG‑H and CH‑R. Data Curation: SR‑C**,** RL‑R**,** LR‑R**,** NG‑H**,** CH‑R and JGR‑A. Writing ‑ Original Draft: SR‑C**,** RL‑R**,** LR‑R**,** GD**,** KB**,** NG‑H**,** AD**,** LA**,** FM‑C and MAC‑J. Writing ‑ Review & Editing: SR‑C**,** RL‑R**,** LR‑R**,** GD**,** KB**,** NG‑H**,** AD**,** LA**,** FM‑C**,** CH‑R and MAC‑J. Visualization: SR‑C**,** RL‑R**,** CH‑R and MAC‑J. Supervision: SR‑C**,** RL‑R**,** LR‑R and CH‑R. Project administration: SR‑C and RL‑R. Funding: RL‑R. Acknowledgements We gratefully acknowledge the advice and support of Dr. José-Luis Jiménez and Dr. Patricia Ripoll of the Aireamos Group, and the funding by generous research grants from the Consejo Potosino de Ciencia y Tecnología (FME/2021/SO-02/14), and the Balvi Clean Air Initiative (A16). References Cincinelli, A. & Martellini, T. Indoor Air Quality and Health. Int. J. Environ. Res. Public. Health . 14 , 1286 (2017). Honan, D., Gallagher, J., Garvey, J. & Littlewood, J. Indoor Air Quality in Naturally Ventilated Primary Schools: A Systematic Review of the Assessment & Impacts of CO2 Levels. Buildings 14 , 4003 (2024). Shukla, A., Indaliya, R. & Tandel, B. N. Effect of Indoor Air Quality on Respiratory Health of Children: An MPPD Model Approach. Aerosol Sci. Eng. 9 , 180–194 (2025). Ge, B. et al. Decision tools for schools using continuous indoor air quality monitors: a case study of CO2 in Boston Public Schools. Lancet Reg. Health - Am. 48 , 101148 (2025). Pérez Velasco, R. & Jarosińska, D. Update of the WHO global air quality guidelines: Systematic reviews – An introduction. Environ. Int. 170 , 107556 (2022). Setty, R. et al. Design Guidance for Education Facilities: Prioritization for Advanced Indoor Air Quality Version 2.0. (2023). US EPA. The EPA’s National School Indoor Air Quality Newsletter - Fall. (2024). https://content.govdelivery.com/accounts/USEPAIAQ/bulletins/3bd7230 (2024). Lv, M., Liu, S., Cao, Q., Zhang, T. & Liu, J. Influence of Ventilation on Indoor Air Quality. in Handbook of Indoor Air Quality (eds Zhang, Y., Hopke, P. K. & Mandin, C.) 1–38 (Springer Singapore, Singapore, doi: 10.1007/978-981-10-5155-5_54-1 . (2022). Peng, Z. & Jimenez, J. L. Exhaled CO 2 as a COVID-19 Infection Risk Proxy for Different Indoor Environments and Activities. Environ. Sci. Technol. Lett. 8 , 392–397 (2021). CDC. Ventilation Can Reduce Exposure to Respiratory Viruses in Indoor Spaces. National Center for Immunization and Respiratory Diseases (2025). https://www.cdc.gov/ncird/whats-new/ventilation-respiratory-viruses.html Zaniboni, L. & Albatici, R. Natural and Mechanical Ventilation Concepts for Indoor Comfort and Well-Being with a Sustainable Design Perspective: A Systematic Review. Buildings 12 , 1983 (2022). Mustafa, M., Cook, M. J. & McLeod, R. S. A critical review of ventilation effectiveness in naturally ventilated spaces from the perspective of sustainability and health. Build. Environ. 270 , 112471 (2025). Becerra, J. A. et al. Identification of potential indoor air pollutants in schools. J. Clean. Prod. 242 , 118420 (2020). Vargas-Legarda, Y. A., Toro-Martínez, A. K. & Rojas-Roa, N. Y. Fajardo-Montaña, O. A. School Children’s exposure to 〖PM〗_(2.5) in a high pollution area of Bogotá, Colombia. Ing. E Investig . 43 , e96125 (2023). Martínez, L. et al. Indoor-outdoor concentrations of fine particulate matter in school building microenvironments near a mine tailing deposit. AIMS Environ. Sci. 3 , 752–764 (2016). Diaz, M., Cools, M., Trebilcock, M., Piderit-Moreno, B. & Attia, S. Effects of Climatic Conditions, Season and Environmental Factors on CO2 Concentrations in Naturally Ventilated Primary Schools in Chile. Sustainability 13 , 4139 (2021). Sahin, C., Dogan, R., Yildiz, T. & Sofuoglu, S. C. M. Indoor environmental quality in naturally ventilated schools of a dusty region: Excess health risks and effect of heating and desert dust transport. Indoor Air 32 , (2022). Gil-Baez, M. et al. Natural ventilation in classrooms for healthy schools in the COVID era in Mediterranean climate. Build. Environ. 206 , 108345 (2021). Oertel, M. et al. Drought Propagation in Semi-Arid River Basins in Latin America: Lessons from Mexico to the Southern Cone. Water 10 , 1564 (2018). Rose, O. G., Bousiotis, D., Rathbone, C. & Pope, F. D. Investigating Indoor Air Pollution Sources and Student’s Exposure Within School Classrooms: Using a Low-Cost Sensor and Source Apportionment Approach. Indoor Air 5544298 (2024). (2024). Higgins, C., Kumar, P. & Morawska, L. Indoor air quality monitoring and source apportionment using low-cost sensors. Environ. Res. Commun. 6 , 012001 (2024). Apostolopoulos, I. D. et al. Monitoring of Indoor Air Quality in a Classroom Combining a Low-Cost Sensor System and Machine Learning. Chemosensors 13 , 148 (2025). INEGI. Directorio Nacional de Unidades Económicas. DENUE. Censos Económicos 2014 (2014). https://www.inegi.org.mx/app/mapa/denue/default.aspx Allen, J., Spengler, J., Jones, E. & Cedeno-Laurent, J. 5-step guide to checking ventilation rates in classrooms. (2020). https://schools.forhealth.org/wp-content/uploads/sites/19/2020/10/Harvard-Healthy-Buildings-program-How-to-assess-classroom-ventilation-10-30-2020.pdf-EN.pdf Allen, J. G. et al. Associations of Cognitive Function Scores with Carbon Dioxide, Ventilation, and Volatile Organic Compound Exposures in Office Workers: A Controlled Exposure Study of Green and Conventional Office Environments. Environ. Health Perspect. 124 , 805–812 (2016). Chen, T., Guestrin, C. & XGBoost: A Scalable Tree Boosting System. in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining 785–794ACM, San Francisco California USA, (2016). 10.1145/2939672.2939785 Géron, A. Hands-on Machine Learning with Scikit-Learn, Keras, and TensorFlow . (2019). Kuhn, M. & Johnson, K. Applied Predictive Modeling (Springer New York, 2013). 10.1007/978-1-4614-6849-3 Hyndman, R. J. & Athanasoupulos, G. Forecasting: Principles and Practice (3rd Ed) OTexts, Melbourne, Australia,. (2021). Rudnick, S. N. & Milton, D. K. Risk of indoor airborne infection transmission estimated from carbon dioxide concentrationAbstract. Indoor Air . 13 , 237–245 (2003). Bazant, M. Z. & Bush, J. W. M. A guideline to limit indoor airborne transmission of COVID-19. Proc. Natl. Acad. Sci. 118, e2018995118 (2021). Zhang, M. et al. The effect of using phase change materials in solar panel cooling to provide green and sustainable energy of a building. J. Build. Eng. 57 , 104933 (2022). Cai, Q. et al. Evaluation of Low-Cost CO2 Sensors Using Reference Instruments and Standard Gases for Indoor Use. Sensors 24 , 2680 (2024). Park, D., Yoo, G. W., Park, S. H. & Lee, J. H. Assessment and Calibration of a Low-Cost PM2.5 Sensor Using Machine Learning (HybridLSTM Neural Network): Feasibility Study to Build an Air Quality Monitoring System. Atmosphere 12, 1306 (2021). Sadrizadeh, S. et al. Indoor air quality and health in schools: A critical review for developing the roadmap for the future school environment. J. Build. Eng. 57 , 104908 (2022). Abdullah, H. K. & Alibaba, H. Z. A Performance-Based Window Design and Evaluation Model for Naturally Ventilated Offices. Buildings 12 , 1141 (2022). Ragab, A., Hassieb, M. M. & Mohamed, A. F. Exploring the impact of window design and ventilation strategies on air quality and thermal comfort in arid educational buildings. Sci. Rep. 15 , 19596 (2025). Garcia, M. S., Souza, R. V. G. & Silva, F. J. Towards Health and Comfort: Architectural Design and Indoor Air Quality in Naturally Ventilated Classrooms. in Proceedings of CIRMARE 2023 (eds. Lanzinha, J. C. G. & Qualharini, E. L.) vol. 444 403–413Springer Nature Switzerland, Cham, (2024). Lawal, A. A., Dare-Abel, O. A. & Ajayi, O. O. A Review of Indoor Environmental Quality (Ieq) Parameters and Their Impact on Occupant Satisfaction in Educational Buildings. 9 , (2025). Fisk, W. J. The ventilation problem in schools: literature review. Indoor Air . 27 , 1039–1051 (2017). Settimo, G. et al. CO2 Levels in Classrooms: What Actions to Take to Improve the Quality of Environments and Spaces. Sustainability 16 , 8619 (2024). Badyda, A. & Rogulski, M. Closing Access Streets to Schools for Vehicular Traffic—Does. It Affect. Air Quality? Sustainability . 16 , 3380 (2024). Alameddine, I., Gebrael, K., Hanna, F. & El-Fadel, M. Quantifying indoor PM2.5 levels and its sources in schools: What role does location, chalk use, and socioeconomic equity play ? Atmospheric Pollut Res. 13 , 101375 (2022). Alves, C., Nunes, T., Silva, J. & Duarte, M. Comfort Parameters and Particulate Matter (PM10 and PM2.5) in School Classrooms and Outdoor Air. Aerosol Air Qual. Res. 13 , 1521–1535 (2013). Pan, H. & Wu, C. Bayesian optimization + XGBoost based life cycle carbon emission prediction for residential buildings—An example from Chengdu, China. Build. Simul. 16 , 1451–1466 (2023). Zhang, R., Li, Y., Zhang, A. L., Wang, Y. & Molina, M. J. Identifying airborne transmission as the dominant route for the spread of COVID-19. Proc. Natl. Acad. Sci. 117, 14857–14863 (2020). Shah, K. B. et al. Evaluating Indoor Air Quality in Residential Environments: A Study of PM2.5 and CO2 Dynamics Using Low-Cost Sensors. Environments 11 , 237 (2024). Dai, Z., Yuan, Y., Zhu, X. & Zhao, L. A Method for Predicting Indoor CO2 Concentration in University Classrooms: An RF-TPE-LSTM Approach. Appl. Sci. 14 , 6188 (2024). Dai, Y. et al. Synergistic effects of CO2 and air pollutants from ship emissions in Shanghai, China: Spatial-temporal characteristics, prediction assessment, policy implications. J. Environ. Manage. 376 , 124417 (2025). Yu, W. et al. Indoor PM2.5 forecasting and the association with outdoor air pollution: a modelling study based on sensor data in Australia. Preprint at. https://doi.org/10.48550/arXiv.2405.07404 (2024). Elbayoumi, M. & Ramli, N. A. Fitri Md Yusof, N. F. Development and comparison of regression models and feedforward backpropagation neural network models to predict seasonal indoor PM2.5–10 and PM2.5 concentrations in naturally ventilated schools. Atmospheric Pollut Res. 6 , 1013–1023 (2015). WHO. WHO Global Air Quality Guidelines: Particulate Matter (PM2.5 and PM10), Ozone, Nitrogen Dioxide, Sulfur Dioxide and Carbon Monoxide (World Health Organization, 2021). Chaudhry, S. K. et al. Influence of seasonal variation on spatial distribution of PM2.5 concentration using low-cost sensors. Environ. Monit. Assess. 196 , 1234 (2024). Elbayoumi, M. et al. Multivariate methods for indoor PM10 and PM2.5 modelling in naturally ventilated schools buildings. Atmos. Environ. 94 , 11–21 (2014). Buonanno, G., Stabile, L. & Morawska, L. Estimation of airborne viral emission: Quanta emission rate of SARS-CoV-2 for infection risk assessment. Environ. Int. 141 , 105794 (2020). Vouriot, C. V. M., Burridge, H. C., Noakes, C. J. & Linden, P. F. Seasonal variation in airborne infection risk in schools due to changes in ventilation inferred from monitored carbon dioxide. Indoor Air . 31 , 1154–1163 (2021). Maantay, J. Mapping environmental injustices: pitfalls and potential of geographic information systems in assessing environmental health and equity. Environ. Health Perspect. 110 , 161–171 (2002). Chakraborty, J. & Maantay, J. A. Proximity Analysis for Exposure Assessment in Environmental Health Justice Research. in Geospatial Analysis of Environmental Health (eds Maantay, J. A. & McLafferty, S.) 111–138 (Springer Netherlands, Dordrecht, doi: 10.1007/978-94-007-0329-2_5 . (2011). US EPA, O. Reference Guide for Indoor Air Quality in Schools. (2014). https://www.epa.gov/iaq-schools/reference-guide-indoor-air-quality-schools Matthaios, V. N. et al. The effects of urban green space and road proximity to indoor traffic-related PM2.5, NO2, and BC exposure in inner-city schools. J. Expo Sci. Environ. Epidemiol. 34 , 745–752 (2024). INECC. Evaluación de Contaminantes Atmosféricos en San Luis Potosí . 32 (2018). https://www.gob.mx/cms/uploads/attachment/file/399489/Informe_Ejecutivo_SLP_Rev3__1_.pdf Al Osman, M., Yang, F. & Massey, I. Y. Exposure routes and health effects of heavy metals on children. BioMetals 32 , 563–573 (2019). Bora, J. et al. IGI Global Scientific Publishing,. Health Effects of Heavy Metals Contamination in Children. in Nanotechnology Applications and Innovations for Improved Soil Health 254–275 (2024). 10.4018/979-8-3693-1471-5.ch012 Gartland, N. et al. The Effects of Traffic Air Pollution in and around Schools on Executive Function and Academic Performance in Children: A Rapid Review. Int. J. Environ. Res. Public. Health . 19 , 749 (2022). Tables Table 1. Ventilation features of classrooms and other rooms Classrooms (N = 73) Features Mean Std Min 25% 50% 75% Max Min ACH 9.81 3.68 3.00 7.30 9.20 11.90 18.90 Max ACH (m) 14.90 5.73 3.54 10.94 14.74 18.44 29.28 Average ACH (m) 22.14 11.01 1.00 13.00 18.30 34.00 40.20 Score 2.85 1.83 0.98 1.70 2.40 3.30 12.12 Other rooms (N = 33) Features Mean Std Min 25% 50% 75% Max Min ACH 6.92 3.55 1.80 5.00 6.10 8.60 16.80 Max ACH (m) 11.62 6.53 2.02 7.20 9.42 15.36 29.52 Average ACH (m) 17.76 11.65 2.50 9.30 12.50 28.30 41.00 Score 1.94 1.15 0.65 1.07 1.63 2.38 4.87 Key: Air changes per hour (ACH). A score lower than one indicates poor ventilation . A score between one and two indicates acceptable ventilation . A score greater than two indicates excellent ventilation . Additional Declarations No competing interests reported. Supplementary Files SupplementaryMaterial.docx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 23 Dec, 2025 Editor assigned by journal 20 Dec, 2025 Submission checks completed at journal 20 Dec, 2025 First submitted to journal 16 Dec, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8379552","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":562600505,"identity":"994c0729-924a-4481-8c75-80c8643b4973","order_by":0,"name":"Salvador Ruiz-Correa","email":"","orcid":"","institution":"Instituto Potosino de Investigación Científica y Tecnológica","correspondingAuthor":false,"prefix":"","firstName":"Salvador","middleName":"","lastName":"Ruiz-Correa","suffix":""},{"id":562600506,"identity":"4394d17a-0042-4642-9b3c-1337f5606822","order_by":1,"name":"Rubén López-Revilla","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA00lEQVRIiWNgGAWjYDACdjBpw8PADOEzNhDUAlGZBtFygAQthyEcorSYNzMfe/Dhz3kZc3bmZ9IfGGxkNxzgTpPAp0XmMFu64Qye2zyWzWxmEgcY0ow3HODdbIBPiwQzj5k0j8RtHoPDPGxALYcTgVo2PiCo5Y/BOZiW/yAtGw4Q1MKQcACm5QAxtgD90nMgGeQXY4szBsnGMw8T8gt787EHP/7Y2ZvzH354o6LCTrbveO82vCEGBGxgEmgwiwSIhCUDorQwfyCseBSMglEwCkYiAAAN2kKwVBRtjQAAAABJRU5ErkJggg==","orcid":"","institution":"Instituto Potosino de Investigación Científica y Tecnológica","correspondingAuthor":true,"prefix":"","firstName":"Rubén","middleName":"","lastName":"López-Revilla","suffix":""},{"id":562600507,"identity":"719e0bdf-3a5e-4d28-b0ee-e44631ffacf7","order_by":2,"name":"Lina Riego-Ruiz","email":"","orcid":"","institution":"Instituto Potosino de Investigación Científica y Tecnológica","correspondingAuthor":false,"prefix":"","firstName":"Lina","middleName":"","lastName":"Riego-Ruiz","suffix":""},{"id":562600508,"identity":"2f4d26e2-3721-4eb7-b9ed-6491cb896c47","order_by":3,"name":"Gladys Díaz","email":"","orcid":"","institution":"Université Sorbonne Paris Nord","correspondingAuthor":false,"prefix":"","firstName":"Gladys","middleName":"","lastName":"Díaz","suffix":""},{"id":562600513,"identity":"935b9586-bf93-41ec-aa0d-fb820dda3bf5","order_by":4,"name":"Khaled Boussetta","email":"","orcid":"","institution":"Université Sorbonne Paris Nord","correspondingAuthor":false,"prefix":"","firstName":"Khaled","middleName":"","lastName":"Boussetta","suffix":""},{"id":562600514,"identity":"abe26f61-2633-4dfa-9d0e-dd9bf1d945ef","order_by":5,"name":"Nicolás Gómez-Hernández","email":"","orcid":"","institution":"Instituto Potosino de Investigación Científica y Tecnológica","correspondingAuthor":false,"prefix":"","firstName":"Nicolás","middleName":"","lastName":"Gómez-Hernández","suffix":""},{"id":562600515,"identity":"fa83afed-d993-4e69-b540-283d848b47a1","order_by":6,"name":"Abdulhalim Dandoush","email":"","orcid":"","institution":"University of Doha for Science and Technology","correspondingAuthor":false,"prefix":"","firstName":"Abdulhalim","middleName":"","lastName":"Dandoush","suffix":""},{"id":562600516,"identity":"82bcf9d8-7626-42d7-a71b-4d1458385258","order_by":7,"name":"Lamine Amour","email":"","orcid":"","institution":"ESME Engineering School","correspondingAuthor":false,"prefix":"","firstName":"Lamine","middleName":"","lastName":"Amour","suffix":""},{"id":562600517,"identity":"3e99bf43-0de9-4189-b15f-a494fdb75027","order_by":8,"name":"Francisco Marmolejo-Cossío","email":"","orcid":"","institution":"Boston College","correspondingAuthor":false,"prefix":"","firstName":"Francisco","middleName":"","lastName":"Marmolejo-Cossío","suffix":""},{"id":562600519,"identity":"f4e40d67-3ece-4aab-a70f-bdad1d904c5e","order_by":9,"name":"Cresencio Hernández-Rosales","email":"","orcid":"","institution":"Instituto Potosino de Investigación Científica y Tecnológica","correspondingAuthor":false,"prefix":"","firstName":"Cresencio","middleName":"","lastName":"Hernández-Rosales","suffix":""},{"id":562600521,"identity":"69890120-3374-4e32-a8e6-e9da61c26193","order_by":10,"name":"Marco Aurelio Cárdenas-Juárez","email":"","orcid":"","institution":"Universidad Autónoma de San Luis Potosí","correspondingAuthor":false,"prefix":"","firstName":"Marco","middleName":"Aurelio","lastName":"Cárdenas-Juárez","suffix":""},{"id":562600523,"identity":"0c10889a-7d6b-4f30-9afe-ea90182ff22f","order_by":11,"name":"Janeth Gabriela Rivera-Aguilar","email":"","orcid":"","institution":"Université Sorbonne Paris Nord","correspondingAuthor":false,"prefix":"","firstName":"Janeth","middleName":"Gabriela","lastName":"Rivera-Aguilar","suffix":""}],"badges":[],"createdAt":"2025-12-16 20:08:33","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8379552/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8379552/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":98888042,"identity":"576c37fa-a1ab-4f58-8dc5-c43a4ba0752b","added_by":"auto","created_at":"2025-12-23 15:31:12","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4028028,"visible":true,"origin":"","legend":"","description":"","filename":"MSAirSentinel.docx","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/c8d64fe7554017e4c5b78390.docx"},{"id":99309462,"identity":"4b5743b0-44aa-401f-99b5-8b51b501f752","added_by":"auto","created_at":"2025-12-31 16:10:26","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":12538,"visible":true,"origin":"","legend":"","description":"","filename":"08a2ce30e63a4d5192247b7472427b12.json","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/2cf10d38c126b40f74a4ca0a.json"},{"id":99309381,"identity":"2050fa36-cac7-4566-a0c2-2187e5966dbb","added_by":"auto","created_at":"2025-12-31 16:10:18","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":13292988,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/e5b94eb79ba83985e5911ca3.docx"},{"id":98888041,"identity":"319fa538-539a-4b74-8736-79aaaf766a59","added_by":"auto","created_at":"2025-12-23 15:31:12","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":149779,"visible":true,"origin":"","legend":"","description":"","filename":"08a2ce30e63a4d5192247b7472427b121enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/363a86babaef03f5b866f442.xml"},{"id":98888051,"identity":"90c2fbf3-845d-4607-910f-c1f1dec6333d","added_by":"auto","created_at":"2025-12-23 15:31:13","extension":"emf","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":90493740,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.emf","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/83ee9947d7e2c7b32983031e.emf"},{"id":99309304,"identity":"03ff8e28-61a2-4126-abb6-6077accdff09","added_by":"auto","created_at":"2025-12-31 16:10:03","extension":"jpeg","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":19136,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/2f0806fe07d61d66dc93240d.jpeg"},{"id":98888037,"identity":"3fa28c88-9275-49e6-a318-10bd2bd1bbae","added_by":"auto","created_at":"2025-12-23 15:31:12","extension":"jpeg","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2158,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/887eb76d5c9d5380d8126e59.jpeg"},{"id":98888046,"identity":"c1f90673-9f06-4f06-a665-6f6fa371a09f","added_by":"auto","created_at":"2025-12-23 15:31:13","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":205901,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/f0304ad694fd52642a2613d1.png"},{"id":99309426,"identity":"de704fa4-8620-4c8d-a204-dcb268de74f8","added_by":"auto","created_at":"2025-12-31 16:10:21","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":510272,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/b179077e465f268bca25f6f2.png"},{"id":98888040,"identity":"1f0c5614-9065-460e-9754-9c9348977ff5","added_by":"auto","created_at":"2025-12-23 15:31:12","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3463,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/20c4e57bda2aec23ab717b66.png"},{"id":99309583,"identity":"cf5b30c0-e97a-495d-9b04-7d7fe9ca6df2","added_by":"auto","created_at":"2025-12-31 16:10:46","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1011,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/ab8bf378094c7118b73794d1.png"},{"id":98888043,"identity":"302d55b3-8f01-4f99-af55-33e8b0b57783","added_by":"auto","created_at":"2025-12-23 15:31:13","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":39085,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/ed3769044c0a196e719069de.png"},{"id":99309314,"identity":"c62f3904-4f4d-4b60-afab-709a4652a382","added_by":"auto","created_at":"2025-12-31 16:10:05","extension":"xml","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":145962,"visible":true,"origin":"","legend":"","description":"","filename":"08a2ce30e63a4d5192247b7472427b121structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/0444e09c9e92f719b370be80.xml"},{"id":98888047,"identity":"e4b73e21-5ee5-4b28-9aa3-ac47ffd3261e","added_by":"auto","created_at":"2025-12-23 15:31:13","extension":"html","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":163216,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/f28c10e13082689abae89563.html"},{"id":98888034,"identity":"5a54f8f6-1bfa-4f43-b10b-c0c283db7d70","added_by":"auto","created_at":"2025-12-23 15:31:12","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":125602,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSpatial distribution of 826 air pollution sources across the San Luis Potosí Metropolitan Area (SLPMA) mapped using data from Instituto Nacional de Estadística y Geografía.\u003c/strong\u003e Emission sources —represented as dots— include paint and chemical stores, incinerators, landfills, brick kilns, gas stations, carpentry workshops, mining operations, cement plants, sugar mills, and high-traffic zones such as supply centers, promenades, and bus terminals. Sources with available quantitative emission data are marked with triangles. Spatial clustering of pollution sources is visualized through kernel density contours generated using Silverman’s method, highlighting areas of concentrated environmental pressure.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/bb5d0b73434286951d93bd9c.png"},{"id":99309387,"identity":"71e22160-88b2-4fbf-9387-66de1eb1dbea","added_by":"auto","created_at":"2025-12-31 16:10:18","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":86399,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePopulation density (inhabitants per square kilometer) across the SLPMA.\u003c/strong\u003e Colored stars indicate the locations of the schools included in our sample. The map polygons correspond to statistical geographic units (AGEBs) defined by INEGI. Population density is classified into five categories: \u003cem\u003eVery high\u003c/em\u003e (23,504–29,378), \u003cem\u003eHigh\u003c/em\u003e (17,628–23,503), \u003cem\u003eMedium\u003c/em\u003e(11,753–17,627), \u003cem\u003eLow\u003c/em\u003e (5,877–11,752), and \u003cem\u003eVery low\u003c/em\u003e (1–5,876).\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/800c364912614ec74338d79d.png"},{"id":98888036,"identity":"1c9c8200-83e4-40b8-9129-ea892f5d469d","added_by":"auto","created_at":"2025-12-23 15:31:12","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":69304,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAverage CO₂-based infection probability for susceptible individuals across all classrooms during class hours, in the hot season (A), and the cold season (B). \u003c/strong\u003eIn both cases, infection probability increased exponentially over time, with a steeper rise observed during the cold season due to elevated CO₂ concentrations linked to reduced ventilation.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/9df723e376da161ec53e6cba.png"},{"id":99322685,"identity":"8ea695f7-c8cc-41cd-b45e-37f37527079a","added_by":"auto","created_at":"2025-12-31 16:43:58","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1526416,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/1da975e0-f959-428c-b9df-38ad31df7fb6.pdf"},{"id":98888050,"identity":"a49e9998-7a2d-40e7-a763-fb233f25c891","added_by":"auto","created_at":"2025-12-23 15:31:13","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":13292988,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-8379552/v1/daea9118fdeeccdc11e32e00.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Air Sentinel: An IoT-Based Platform for Monitoring Indoor Air Quality in Elementary Schools of the Global South","fulltext":[{"header":"Introduction","content":"\u003cp\u003eIn recent years, numerous studies have shown that indoor air can pose greater health risks than outdoor air.\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e In elementary schools, indoor air quality (IAQ) is a critical yet often overlooked factor influencing children\u0026rsquo;s respiratory health.\u003csup\u003e\u003cspan additionalcitationids=\"CR3\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e Classrooms frequently harbor a range of indoor air pollutants, including carbon dioxide (CO₂), fine particulate matter (PM\u003csub\u003e2.5\u003c/sub\u003e), volatile organic compounds, and allergens, all of which can impact children's health and learning. International guidelines and standards, such as those of the World Health Organization and the American Society of Heating, Refrigerating, and Air-Conditioning Engineers, define thresholds for indoor pollutants that many schools fail to meet.\u003csup\u003e\u003cspan additionalcitationids=\"CR6\" citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eCO\u003csub\u003e2\u003c/sub\u003e and PM\u003csub\u003e2.5\u003c/sub\u003e are key indoor pollutants closely tied to ventilation quality and respiratory health.\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e,\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e Elevated CO₂ levels, mainly from human respiration in crowded or poorly ventilated spaces, signal inadequate air exchange and are linked to reduced student cognitive performance\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e and increased airborne infection risk.\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e Poor ventilation allows pathogens like influenza and SARS-CoV-2 to accumulate, while effective natural or mechanical systems improve IAQ by diluting contaminants.\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e In classrooms, proper airflow not only reduces the transmission of aerosols but also enhances thermal comfort.\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e,\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e PM\u003csub\u003e2.5\u003c/sub\u003e, originating from outdoor sources, cleaning, and resuspension, also poses serious health risks, especially for children with developing lungs and immune systems.\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e,\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e Exposure is linked to asthma, bronchitis, impaired lung function, and long-term cardiovascular issues. Together, CO₂ and PM\u003csub\u003e2.5\u003c/sub\u003e are vital IAQ indicators, and their systematic monitoring supports targeted ventilation strategies and broader interventions for healthier classroom environments.\u003c/p\u003e \u003cp\u003eStudies in the Global South have employed continuous monitoring and cross-sectional sampling to assess school IAQ. In Colombia, PM\u003csub\u003e2.5\u003c/sub\u003e data paired with activity logs and respiratory surveys showed that younger children inhale higher pollutant doses, often exceeding safe limits and reporting more symptoms.\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e Chilean schools with natural ventilation recorded CO₂ levels regularly above 1400 ppm, peaking at 5000 ppm in winter due to seasonal, architectural, and occupancy factors.\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e,\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e In Turkey, classrooms exhibited median CO₂ levels above 1000 ppm and PM\u003csub\u003e2.5\u003c/sub\u003e averages of 31.8 \u0026micro;g/m\u0026sup3;, which correlated with increased respiratory symptoms and an increased risk of bronchitis.\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e Research from Portugal and Spain found that classroom activities, such as cleaning and window use, can cause sharp, short-term spikes in PM2.5.\u003csup\u003e18\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eDespite growing global attention to IAQ, Mexican cities, particularly the San Luis Potos\u0026iacute; Metropolitan Area (SLPMA), remain underrepresented in research and policy. Situated in a dusty, semi-arid region,\u003csup\u003e19\u003c/sup\u003e SLPMA\u0026rsquo;s elementary schools face compounded IAQ risks from rapid urbanization, industrial emissions (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e), and deteriorating infrastructure. These conditions threaten children's respiratory health and academic outcomes; however, the absence of systematic monitoring hampers the identification of high-risk classrooms and the development of targeted interventions.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eExpanding IAQ research in cities like SLPMA is crucial for safeguarding vulnerable populations and guiding context-specific solutions. While active and passive sampling methods have been used for over two decades, the rise of low-cost online sensors offers new, affordable avenues for IAQ monitoring,\u003csup\u003e20\u0026ndash;22\u003c/sup\u003e which is a vital development for the Global South, where financial constraints often limit access to conventional monitoring systems.\u003c/p\u003e \u003cp\u003eThis study evaluated ventilation and IAQ across 106 spaces, including classrooms and other rooms, within 53 of the 315 public elementary schools in the SLPMA.\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e By analyzing CO₂ and PM\u003csub\u003e2.5\u003c/sub\u003e concentrations, the research aimed to identify significant risk factors and offer evidence-based recommendations to improve health and safety in learning environments. A set of targeted research questions guided the study:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eHow adequate is classroom ventilation in public elementary schools in San Luis Potos\u0026iacute;?\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eHow do CO₂ and PM\u003csub\u003e2.5\u003c/sub\u003e levels fluctuate across seasons in classrooms?\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eHow do seasonal dynamics in indoor CO₂ and PM\u003csub\u003e2.5\u003c/sub\u003e concentrations across classrooms shape the reliability and generalizability of data-driven forecasting machine learning models?\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eTo address these questions, this study presents a scalable model for low-cost IAQ monitoring in resource-limited settings, using the SLPMA as a case example. By addressing seasonal patterns, pollutant forecasting, and ventilation strategies, it responds to structural and climatic challenges common across emergent economies in the Global South. The integration of real-time data and predictive modeling enables local stakeholders to enhance children's health and learning conditions, thereby promoting environmental justice and public health equity.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eSampled Schools\u003c/h2\u003e \u003cp\u003eOur study sample comprises 53 elementary public schools selected by the local education ministry, with a focus on those located in areas with high COVID-19 incidence during the pandemic. Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e shows population density across the SLPMA, with colored stars marking sampled schools and polygons representing statistical geographic units (AGEBs) defined by Instituto Nacional de Estad\u0026iacute;stica y Geograf\u0026iacute;a (INEGI).\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e Among these schools, 14.0% are located in very high-density areas, 18.7% in high-density areas, 33.3% in medium-density areas, 27.0% in low-density areas, and 6.4% in very low-density areas. Figure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e displays the Marginalization Index (MI) across the SLPMA, developed by the Consejo Nacional de Poblaci\u0026oacute;n to measure social and economic deprivation using indicators in education, housing, income, and population distribution.\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e Colored stars mark sampled schools, with 50.0% in very low marginalization areas, 37.5% in low, and 12.5% in medium marginalization zones.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eAir Sentinel\u003c/b\u003e \u003cb\u003eDeployment\u003c/b\u003e\u003c/p\u003e \u003cp\u003e \u003cem\u003eAir Sentinel\u003c/em\u003e was installed with parental consent in 11 schools to continuously monitor CO₂, PM\u003csub\u003e2.5\u003c/sub\u003e, temperature, and humidity (Figure S2). One classroom was selected per school, and data were collected during four epidemiological weeks in 2023: weeks 23 and 25 (hot season, 13\u0026ndash;30 \u003csup\u003eo\u003c/sup\u003eC), 46 and 48 (cold season, 5\u0026ndash;14 \u003csup\u003eo\u003c/sup\u003eC). Although limited connectivity affected data quality in weeks 25 and 46, CO₂ and PM\u003csub\u003e2.5\u003c/sub\u003e were successfully captured across all four weeks. Primary analyses use complete datasets from weeks 23 and 48, while CO₂ and PM\u003csub\u003e2.5\u003c/sub\u003e forecasting incorporates.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eGeometric Features and Ventilation Metrics of Classrooms\u003c/h3\u003e\n\u003cp\u003eThe study assessed 73 classrooms and 33 additional rooms (e.g., libraries, labs) across 53 schools, using nine descriptors. Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e describes the geometric features of the 106 spaces: most had rectangular bases (only one was pentagonal). In all spaces, the window area exceeded 10% of the floor area, and windows were evenly split between sliding and top-hinged types, meeting lighting and ventilation standards. Maintenance issues affected 56% of the spaces; 39.2% lacked cross-ventilation. Occupancy density averaged 1.13\u0026thinsp;\u0026plusmn;\u0026thinsp;0.33 per m\u0026sup2; in classrooms and 1.07\u0026thinsp;\u0026plusmn;\u0026thinsp;0.37 in other rooms, both of which are above the legal minimum of 0.625.\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;1 presents ventilation data for the 106 spaces. Forced ventilation was more common in classrooms (68.5%) than in other rooms (31.5%). IAQ ratings showed disparities: poor in 27.3% of the classrooms compared to 1.3% in other rooms, and excellent in 67.2% compared to 34.6% in other rooms. Acceptable IAQ ratings were similar for classrooms (31.5%) and other rooms (36.4%). Figure S3 shows the variability in ventilation scores (box plots). The magnitude of scores and IAQ levels of all school rooms is mapped in Figures S4 and S5 using scaled, shaded circles.\u003c/p\u003e\n\u003ch3\u003eCO and PM Concentration Metrics\u003c/h3\u003e\n\u003cp\u003eCO₂ and PM\u003csub\u003e2.5\u003c/sub\u003e were continuously monitored in 11 schools during class hours (08:00\u0026ndash;14:00 h) using \u003cem\u003eAir Sentinel\u003c/em\u003e. Table S2 summarizes the seasonal variation in CO₂ and PM\u003csub\u003e2.5\u003c/sub\u003e concentrations across monitored school spaces (Figure S6). CO₂ levels were higher during the cold season, with a mean of 894.45 ppm (CV\u0026thinsp;=\u0026thinsp;0.44), compared to 666.67 ppm (CV\u0026thinsp;=\u0026thinsp;0.21) in the hot season. The cold season also exhibited greater dispersion, with a maximum value of 3069.85 ppm, more than double that of the hot season (1226.38 ppm), and a wider interquartile range (588.44\u0026ndash;1105.44 ppm).\u003c/p\u003e \u003cp\u003ePM\u003csub\u003e2.5\u003c/sub\u003e concentrations showed an inverse seasonal pattern. Mean levels were higher in the hot season (14.66 \u0026micro;g/m\u0026sup3;) than in the cold season (9.67 \u0026micro;g/m\u0026sup3;), and variability was greater in the latter (CV\u0026thinsp;=\u0026thinsp;1.30 vs. 0.70). The cold season also presented extreme outliers, with a maximum of 853.26 \u0026micro;g/m\u0026sup3;, far exceeding typical indoor thresholds and the hot season peak of 58.76 \u0026micro;g/m\u0026sup3;. These findings suggest distinct seasonal dynamics in IAQ, with elevated CO₂ during colder months linked to reduced ventilation, and sporadic PM\u003csub\u003e2.5\u003c/sub\u003e spikes in the cold season, potentially driven by localized sources or infiltration events.\u003c/p\u003e \u003cp\u003ePairwise Pearson\u0026rsquo;s coefficient (PC) for CO₂ time series across classrooms averaged 0.24\u0026thinsp;\u0026plusmn;\u0026thinsp;0.25, with a mean R\u0026sup2; = 0.12, indicating low linear synchrony and predictive power across classrooms. Cold season values were slightly higher (PC\u0026thinsp;=\u0026thinsp;0.26\u0026thinsp;\u0026plusmn;\u0026thinsp;0.24; R\u0026sup2; = 0.13), but still modest.\u003c/p\u003e \u003cp\u003ePM\u003csub\u003e2.5\u003c/sub\u003e analysis showed even weaker linear associations. In the hot season, PC averaged 0.10\u0026thinsp;\u0026plusmn;\u0026thinsp;0.42, and R\u0026sup2; was 0.19; in the cold season; PC dropped to 0.06\u0026thinsp;\u0026plusmn;\u0026thinsp;0.29, and R\u0026sup2; decreased to 0.08, confirming limited linear temporal associations and predictability across classrooms.\u003c/p\u003e\n\u003ch3\u003eForecasting CO and PM Concentrations\u003c/h3\u003e\n\u003cp\u003eDuring the hot season, an XGBoost autoregressive model\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e achieved optimal CO₂ forecasting with an 8-hour lookback window (\u003cem\u003eh\u003c/em\u003e\u0026thinsp;=\u0026thinsp;8). Using leave-one-classroom-out cross-validation, it achieved an average R\u0026sup2; of 0.85, MAE of 12.12 ppm, and RMSE of 17.44 ppm. Generalization gaps were 10.76 ppm (MAE) and 15.63 ppm (RMSE), indicating low overfitting but acceptable robustness for deployment in unseen classrooms.\u003c/p\u003e \u003cp\u003eIn the cold season, the XGBoost model optimal forecasting setup showed slightly reduced performance. Using leave-one-classroom-out cross-validation, the model achieved an average R\u0026sup2; of 0.80, with an MAE of 68.04 ppm and an RMSE of 95.7 ppm, indicating larger prediction deviations. The generalization gaps were 62.95 ppm (MAE) and 88.23 ppm (RMSE), suggesting lower generalizability under colder conditions.\u003c/p\u003e \u003cp\u003eApplying the XGBoost forecasting framework to PM\u003csub\u003e2.5\u003c/sub\u003e concentrations yielded weak results across both hot and cold seasons, with low R\u0026sup2; values and high prediction errors, indicating that short-term PM\u003csub\u003e2.5\u003c/sub\u003e variability cannot be reliably predicted using historical indoor data alone. These findings suggest that PM\u003csub\u003e2.5\u003c/sub\u003e dynamics in schools are driven by external factors such as traffic emissions, cleaning, or resuspension events, which are not captured by the model\u0026rsquo;s input features.\u003c/p\u003e\n\u003ch3\u003eCO Concentration Trends\u003c/h3\u003e\n\u003cp\u003eBeyond aggregate analyses, a detailed assessment of pollutant trends was conducted at the individual classroom level across the 11 schools. Metrics such as the coefficient of variation (CV) and Pearson\u0026rsquo;s coefficient (PC) between CO₂ and PM\u003csub\u003e2.5\u003c/sub\u003e were used to capture intra-room variability and identify classrooms with unstable or elevated profiles. Due to the volume and granularity of the data, classroom-specific statistics are provided in Tables S3\u0026ndash;S6. This deeper analysis enhances understanding of IAQ dynamics and supports targeted interventions for specific classroom conditions.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eCO\u003csub\u003e2\u003c/sub\u003e Concentration Metrics and Infection Probability\u003c/h2\u003e \u003cp\u003eFigure 4 compares average CO₂-based AIP during class hours across schools in the hot (top panel) and cold (bottom panel) seasons. In both cases, AIP values rose exponentially, reaching\u0026thinsp;~\u0026thinsp;0.9 after 150 minutes in the hot season and ~\u0026thinsp;0.95 in the cold season, due to higher CO₂ levels from reduced ventilation. These values were similar across classrooms. Each curve was modeled using \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:y=A(1-{e}^{-Bx})\\)\u003c/span\u003e\u003c/span\u003e (Table S7). A representative case study is illustrated in Figure S7.\u003c/p\u003e \u003cp\u003eXGBoost forecasting showed high predictive accuracy for short-term infection probability during the hot season, using a leave-one-classroom-out strategy to ensure generalizability. It achieved a test R\u0026sup2; of 0.98, with an MAE of 0.005 and an RMSE of 0.008. The generalization gap was minimal (MAE\u0026thinsp;=\u0026thinsp;0.004; RMSE\u0026thinsp;=\u0026thinsp;0.006), indicating strong transferability across diverse classroom conditions.\u003c/p\u003e \u003cp\u003eDuring the cold season, the model maintained robust performance with tighter predictive alignment. Using the leave-one-classroom-out approach, it achieved a test R\u0026sup2; of 0.99, indicating near-perfect correlation between predicted and observed AIP values. Test MAE and RMSE remained at 0.005 and 0.008, with a low generalization gap (MAE\u0026thinsp;=\u0026thinsp;0.004; RMSE\u0026thinsp;=\u0026thinsp;0.006). These results confirm the model\u0026rsquo;s robustness under seasonal IAQ shifts, particularly elevated CO₂ levels resulting from reduced ventilation, without compromising predictive reliability.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eOutdoor PM Pollution Sources\u003c/h3\u003e\n\u003cp\u003eAlthough PM\u003csub\u003e2.5\u003c/sub\u003e concentrations did not reveal consistent statistical patterns across classrooms, the absence of spatial coherence suggests the need to investigate localized pollution and emission sources (for which quantitative data is available). In a complex urban setting like the SLPMA, local PM\u003csub\u003e2.5\u003c/sub\u003e levels are significantly influenced by the surrounding environment. A detailed spatial inventory reveals a diverse array of potential contributors distributed across the city: 143 gas stations, 129 brick kilns, 123 biomass burning sites, 90 carpentry shops, 62 solvent shops, 50 chemical factories, 25 mining operations, 15 incinerators, seven paper factories, seven pesticide factories, six cement factories, six landfills and a concentration of additional emission sources in the southern region. These sources exhibit varying distributions, contributing to air quality heterogeneity (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eA buffer zone analysis was performed using radii ranging from 100 meters to 3 km in 100-meter increments, generating 30 concentric zones around each school to assess the density of nearby pollution sources (Figure S8). For each buffer, pollution sources were quantified using the georeferenced inventory. Additionally, street segments within 100 meters of each school were counted to estimate the potential traffic-related exposure (Figure S9). Sources from nearby streets may emit NO₂, CO, black carbon, and PM\u003csub\u003e2.5\u003c/sub\u003e, especially in congested areas. This spatial analysis serves as a proxy for exposure risk, helping to identify environmental pressure zones, and guide future monitoring studies and mitigation strategies.\u003c/p\u003e \u003cp\u003eTable S4 provides a detailed summary of stationary pollution sources located within a 3 km radius of each elementary school in the SPLMA. Table S5 complements this finding by listing all inventoried emission sources, both stationary and mobile, within the same spatial extent, offering a comprehensive view of the surrounding environmental pressure at the broadest buffer scale considered. A buffer size of 3 km offers a broad spatial context, encompassing not only immediate surroundings but also more distant contributors.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eElementary schools in the SLPMA are predominantly located in medium- and low-density zones (\u0026gt; 60%), indicating concentration in suburban or peri-urban areas. Nearly one-third are in high- or very high-density zones, where environmental stressors such as air pollution and traffic congestion are more intense, while only 6.4% are in very low-density areas, suggesting gaps in semirural access. This pattern aligns with international findings; Sadrizadeh et al.\u003csup\u003e35\u003c/sup\u003e reviewed 304 studies (1970–2022) showing that IAQ disparities often align with urbanization levels, since urban schools exhibiting elevated CO₂, particulate matter, and VOCs levels due to traffic, poor ventilation, and aging infrastructure.\u003c/p\u003e \u003cp\u003eWindow surface areas exceeding 10% of the classroom floor areas meet foundational design standards for natural lighting and ventilation, consistent with ASHRAE and International Residential Code recommendations. However, this does not ensure adequate ventilation, as actual performance depends on the operability and strategic placement of the openings.\u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e While the presence of both sliding and top-hinged windows allows for varied ventilation strategies, 56% of spaces suffer from poor maintenance, compromising window operability and IAQ. Neglected mechanical components can hinder airflow, trap pollutants, and compromise thermal comfort,\u003csup\u003e37\u003c/sup\u003e and in schools where ventilation is vital for cognitive function and infection control, non-functional windows represent lost opportunities for passive environmental regulation.\u003c/p\u003e \u003cp\u003eInsufficient cross-ventilation affects 39.2% of the classrooms, posing a significant barrier to adequate natural ventilation. Cross-ventilation through openings on opposing walls is crucial for enhancing air exchange and reducing pollutants, leading to lower CO₂ levels and better thermal comfort than single-sided ventilation.\u003csup\u003e\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u003c/sup\u003e In San Luis Potosí schools, where mechanical systems may be inconsistently available, architectural solutions are critical. Reported occupancy densities of 1.13 ± 0.33 students/m² in classrooms and 1.07 ± 0.37 persons/m² in other rooms comply with the Mexican legal standard of 0.625 students/m². However, regulatory compliance does not guarantee optimal conditions, as even moderate densities can elevate CO₂ levels if ventilation is inadequate.\u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e Forced ventilation systems are unevenly distributed; present in 68.5% of classrooms but only in 48.5% of other rooms, revealing infrastructure disparities that may reflect focus on instructional areas over administrative ones.\u003csup\u003e\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e Limited coverage in non-classroom areas suggests planning blind spots that could impact staff health and building performance.\u003c/p\u003e \u003cp\u003eOur results revealed significant seasonal variability in IAQ, particularly in CO₂ and PM\u003csub\u003e2.5\u003c/sub\u003e concentrations, closely tied to environmental conditions and ventilation dynamics during class periods (08:00–14:00 hours). CO₂ concentrations were higher during the cold season (mean = 894.45 ppm) compared to the hot season (666.67 ppm) due to reduced natural ventilation from closed windows and doors, with greater variability (CV = 0.44 vs. 0.21). These results align with elevated winter CO₂ levels in naturally ventilated classrooms\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e,\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e Low pairwise Pearson coefficients (PC = 0.24–0.26) and low R² values (0.12–0.13) indicated weak linear temporal synchrony between classrooms, suggesting exploration of non-linear relations using machine learning tools.\u003c/p\u003e \u003cp\u003ePM\u003csub\u003e2.5\u003c/sub\u003e concentrations varied seasonally, with higher mean levels in the hot season (14.66 µg/m³) than the cold season (9.67 µg/m³) due to increased outdoor air exchange. However, the cold season exhibited significantly higher maximum values (853.26 µg/m³) and greater variability (CV = 1.30 vs. 0.70), suggesting episodic spikes from localized sources. This aligns with prior research showing higher summer PM\u003csub\u003e2.5\u003c/sub\u003e from outdoor infiltration while winter peaks were attributed to indoor sources, with chalk, cleaning, socioeconomic factors, and combustion identified as contributors.\u003csup\u003e\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eDuring the hot season, XGBoost achieved strong CO₂ predictive performance with an 8-hour lookback window (R² = 0.85; MAE = 12.12 ppm; RMSE = 17.44 ppm), with moderate generalization gaps suggesting limited overfitting and good robustness.\u003csup\u003e\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e,\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e\u003c/sup\u003e During the cold season, performance declined (R² = 0.80; MAE = 68.04 ppm; RMSE = 95.7 ppm) with generalization gaps indicating reduced adaptability, likely due to variable ventilation behaviors introducing non-linearities.\u003csup\u003e\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e,\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e\u003c/sup\u003e PM\u003csub\u003e2.5\u003c/sub\u003e forecasting yielded poor results in both seasons, with low R² values and high errors, suggesting that levels are driven by complex, short-term fluctuations not captured by historical indoor data. External factors such as traffic emissions, cleaning, and particle resuspension play significant roles,\u003csup\u003e49,50\u003c/sup\u003e emphasizing the need to map surrounding pollution sources.\u003c/p\u003e \u003cp\u003ePM\u003csub\u003e2.5\u003c/sub\u003e data from 11 schools during the spring and winter of 2023 revealed marked seasonal variability.\u003csup\u003e\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e\u003c/sup\u003e In the cold season, several classrooms exceeded WHO's 24-hour threshold of 15 µg/m³,\u003csup\u003e52\u003c/sup\u003e with schools HC, FS, and FN exhibiting repeated high concentrations (HC: 53.1 µg/m³; FS: 34.2 µg/m³) likely due to limited ventilation and proximity to pollution sources. Extreme outliers (JMJ: 853.3 µg/m³; VFS: 456.9 µg/m³) underscore the need for continuous monitoring.\u003csup\u003e\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e\u003c/sup\u003e In the hot season, most classrooms maintained levels below 20 µg/m³, though hotspots like JMJ, HC, and IZ exceeded 25 µg/m³ (peak: 58.4 µg/m³), suggesting persistent challenges.\u003csup\u003e\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eForecasting analysis indicates that CO₂ concentrations can be predicted with moderate accuracy across both seasons, opening the door to CO₂-based modeling of airborne infection risk for respiratory illnesses, as CO₂ reflects occupancy and air exchange. In contrast, PM\u003csub\u003e2.5\u003c/sub\u003e forecasts performed poorly in both seasons, with variability stemming from localized, episodic factors not captured by indoor historical data, emphasizing the need to map external pollution sources.\u003c/p\u003e \u003cp\u003eCO₂-based infection probability analysis reveals clear seasonal patterns with significant implications for IAQ and infection risk management. Infection probability during class hours follows an exponential growth curve in both seasons, with the cold season showing a steeper trajectory due to higher CO₂ levels from reduced ventilation. The exponential saturation model effectively captures these dynamics, and the time to reach 95% infection probability is consistently shorter in the cold season.\u003csup\u003e\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e,\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e\u003c/sup\u003e XGBoost modeling achieved high accuracy in both the hot season (R² = 0.98; MAE = 0.005; RMSE = 0.008) and cold season (R² = 0.99), with strong generalizability confirmed by leave-one-classroom-out strategy, demonstrating adaptability to seasonal IAQ shifts and affirming the feasibility of using CO₂ data for real-time infection risk forecasting.\u003c/p\u003e \u003cp\u003eSpatial buffer analysis using a 3 km radius effectively identifies environmental pressure zones and supports monitoring and policy planning.\u003csup\u003e\u003cspan additionalcitationids=\"CR58 CR59\" citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e–\u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e\u003c/sup\u003e Results reveal an average of 110.1 pollution sources per sampling point, with gas stations, biomass combustion sites, and carpentry shops most common. Brick kilns show extreme variability (mean = 9.5; max = 89), indicating industrial clusters with high particulate emissions.\u003csup\u003e\u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e61\u003c/span\u003e\u003c/sup\u003e High-impact but less frequent sources such as incinerators, cement factories, and landfills may exert disproportionate influence, and wide variability reflects spatial inequality in exposure.\u003csup\u003e\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e\u003c/sup\u003e CO₂ is widespread (138 sources) and often co-occurs with hazardous pollutants such as cadmium, mercury, lead, arsenic, and cyanide,\u003csup\u003e52,62\u003c/sup\u003e with lead standing out (mean = 5.0; max = 21). Even low concentrations of dioxins, furans, and VOCs raise concerns about long-term carcinogenic exposure.\u003csup\u003e\u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e63\u003c/span\u003e\u003c/sup\u003e Schools are closely surrounded by traffic infrastructure (average 8.4 streets within 100 meters), increasing exposure to PM\u003csub\u003e2.5\u003c/sub\u003e, NO₂, and O₃. Elevated NO₂ levels worsen asthma and impair lung function,\u003csup\u003e59,60\u003c/sup\u003e while ground-level ozone threatens respiratory health and contributes to chronic conditions.\u003csup\u003e\u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e64\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eThe \u003cem\u003eAir Sentinel\u003c/em\u003e platform successfully demonstrated the viability of a decentralized, mobile-driven IoT network for IAQ monitoring in resource-limited schools, enabling real-time data collection and visualization across classrooms using Bluetooth, Wi-Fi, and 4G connectivity. However, the limitations included Bluetooth's short range and interference, inconsistent Wi-Fi coverage, cellular signal variability and cost concerns, highlighting the need for more resilient, low-power communication solutions. Despite these constraints, Air Sentinel delivered valuable seasonal IAQ insights and actionable feedback for educators, affirming its scalability for school-based environmental health surveillance.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003cdiv id=\"Sec12\" class=\"Section3\"\u003e \u003c/div\u003e \u003c/div\u003e "},{"header":"Methods","content":"\u003ch2\u003eLocation\u003c/h2\u003e\u003cp\u003eThe SLPMA, comprising San Luis Potosí, Soledad de Graciano Sánchez, and nearby areas, is the state’s most populous and economically vital region. In 2020, its 1.2\u0026nbsp;million residents made it Mexico’s 11th largest metropolitan area, housing nearly one-third of the state’s population. As the leading industrial, commercial, and cultural center, it features strong manufacturing, growing coordination, and a strategic location in Mexico. Public elementary schools in this area are particularly vulnerable to IAQ issues due to the semi-arid climate, rapid urban growth, industrial emissions (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e), and deteriorating infrastructure with inadequate maintenance.\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e\u003cp\u003eVentilation assessments were carried out in 106 spaces across 53 schools in the SLPMA. IAQ measurements were conducted in one classroom per school within a subset of 11 schools, where parents provided signed consent forms authorizing the installation of measurement devices during class hours.\u003c/p\u003e\u003ch2\u003eVentilation Metrics\u003c/h2\u003e\u003cp\u003eWe applied the Harvard Healthy Buildings Program protocol\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e to estimate ventilation schools using the LinkApp mobile tool developed by our research team. Based on CO₂ decay, this method is suitable for naturally or mechanically ventilated areas where direct airflow measurement is challenging. An empty classroom is selected to avoid human-generated CO₂. Room characteristics, including use, size, occupancy, and openings, determine the number of calibrated monitors, which are placed 1–1.5 m above the ground and away from ventilation sources.\u003c/p\u003e\u003cp\u003eDry ice elevates CO₂, which is dispersed with fans until it reaches ~ 2000 ppm. Calibrated monitors (range 0–5000 ppm, accuracy ± 50 ppm) record CO₂ concentration every minute. The rate of decline indicates outdoor air exchange, and each sensor’s decay curve yields its air changes per hour (ACH).\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e The formula assumes a well-mixed indoor environment and constant outdoor CO\u003csub\u003e2\u003c/sub\u003e concentration, and uses the exponential decay model,\u003c/p\u003e\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:ACH\\:=\\:-\\frac{1}{\\varDelta\\:T}\\text{log}\\left(\\frac{{C}_{f}-{C}_{a}}{{C}_{i}-{C}_{a}}\\right),$$\u003c/div\u003e\u003c/div\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003ewhere ACH is expressed in h\u003csup\u003e− 1\u003c/sup\u003e, \u003cem\u003eC\u003c/em\u003e\u003csub\u003e\u003cem\u003ef\u003c/em\u003e\u003c/sub\u003e is the final CO\u003csub\u003e2\u003c/sub\u003e concentration in ppm, \u003cem\u003eC\u003c/em\u003e\u003csub\u003e\u003cem\u003ei\u003c/em\u003e\u003c/sub\u003e is the maximum CO\u003csub\u003e2\u003c/sub\u003e concentration in ppm, \u003cem\u003eC\u003c/em\u003e\u003csub\u003e\u003cem\u003ea\u003c/em\u003e\u003c/sub\u003e is the outdoor CO\u003csub\u003e2\u003c/sub\u003e concentration in ppm, and ∆\u003cem\u003eT\u003c/em\u003e is the time in minutes elapsed at which the decrease in CO\u003csub\u003e2\u003c/sub\u003e concentration is approximately 37% the maximum initial concentration. The ventilation rate is defined as:\u003c/p\u003e\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$\\:Ventilation\\:rate\\:=\\:\\frac{\\:{ACH}_{min}\\times\\:\\:V\\:\\times\\:1000\\:}{3600}\\:\\left(\\frac{L}{s}\\right),$$\u003c/div\u003e\u003c/div\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003ewhere V is the room's volume and \u003cem\u003eACH\u003c/em\u003e\u003csub\u003e\u003cem\u003emin\u003c/em\u003e\u003c/sub\u003e is the lowest ACH measurement obtained from the room monitors. The ventilation score is calculated as the ratio of \u003cem\u003eACH\u003c/em\u003e\u003csub\u003e\u003cem\u003emin\u003c/em\u003e\u003c/sub\u003e to the required ventilation rate (\u003cem\u003eACH\u003c/em\u003e\u003csub\u003e\u003cem\u003ereq\u003c/em\u003e\u003c/sub\u003e), which represents the minimum airflow needed to remove the CO₂ generated by the occupants:\u003c/p\u003e\u003cdiv id=\"Equc\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equc\" name=\"EquationSource\"\u003e\n$$\\:score\\:=\\:\\frac{{ACH}_{min\\:}\\:}{{ACH}_{req}}\\:=\\:\\frac{{CFM}_{min\\:}\\:}{{CFM}_{req}}$$\u003c/div\u003e\u003c/div\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eA score ≥ 2 leads to an excellent IAQ, a score from 1 to 2 leads to an acceptable IAQ, and a score \u0026lt; 1 leads to a poor IAQ.\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e We computed descriptive statistics for \u003cem\u003eACH\u003c/em\u003e\u003csub\u003e\u003cem\u003emin\u003c/em\u003e\u003c/sub\u003e, \u003cem\u003eACH\u003c/em\u003e\u003csub\u003e\u003cem\u003emax\u003c/em\u003e,\u003c/sub\u003e \u003cem\u003eACH\u003c/em\u003e\u003csub\u003e\u003cem\u003eaverage\u003c/em\u003e\u003c/sub\u003e, and the \u003cem\u003escore\u003c/em\u003e in Table\u0026nbsp;2.\u003c/p\u003e\u003ch2\u003eCO\u003csub\u003e2\u003c/sub\u003e and PM\u003csub\u003e2.5\u003c/sub\u003e Concentration Metrics\u003c/h2\u003e\u003csub\u003e2\u003c/sub\u003e\u003cp\u003eWe computed seasonal descriptive statistics, including the coefficient of variation (CV) for CO₂, PM\u003csub\u003e2.5\u003c/sub\u003e, temperature, and humidity across 11 classrooms. We assessed how much variability in one classroom’s CO₂ and PM\u003csub\u003e2.5\u003c/sub\u003e levels could be explained by the levels of another classroom across schools by a linear model.\u003c/p\u003e\u003cp\u003eMean excess CO₂ during class hours was calculated as:\u003c/p\u003e\u003cdiv id=\"Equd\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equd\" name=\"EquationSource\"\u003e\n$$\\:\\varDelta\\:C\\:=\\:\\frac{1}{T}\\:{\\int\\:}_{0}^{T}C\\left(t\\right)\\:dt,$$\u003c/div\u003e\u003c/div\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003ein ppm/min, where \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:C\\left(t\\right)\\)\u003c/span\u003e\u003c/span\u003e denotes the CO₂ instantaneous concentration above 1000 ppm. CO₂ levels of 400–1000 ppm was considered typical, 1000–2000 ppm moderate, and \u0026gt; 2000 ppm high.⁴\u003c/p\u003e\u003ch2\u003eCO\u003csub\u003e2\u003c/sub\u003e and PM\u003csub\u003e2.5\u003c/sub\u003e Forecasting in Hot and Cold Seasons\u003c/h2\u003e\u003csub\u003e2.5\u003c/sub\u003e\u003cp\u003eWe applied a machine learning approach to \u003cem\u003eforecast indoor CO₂ and PM\u003c/em\u003e\u003csub\u003e\u003cem\u003e2.5\u003c/em\u003e\u003c/sub\u003e concentrations one hour into the future, using the Extreme Gradient Boosting (XGBoost) regression. XGBoost is a scalable, regularized gradient boosting framework known for its high predictive accuracy and robustness to overfitting.\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e,\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e\u003cp\u003eThe input dataset comprised CO₂ (PM\u003csub\u003e2.5\u003c/sub\u003e) measurements from 11 naturally ventilated classrooms, each monitored continuously during four nonconsecutive one-week periods, two periods representing the hot season (13°C-30°C) and two periods representing the cold season (5°C-14°C). This seasonal sampling design was intended to capture variability in ventilation behavior and occupancy patterns under contrasting climatic conditions. CO₂ (PM\u003csub\u003e2.5\u003c/sub\u003e) concentrations were initially recorded at 1-minute intervals and subsequently aggregated into 5-minute averages to reduce high-frequency noise and align with operational decision-making times relevant to school environments.\u003c/p\u003e\u003cp\u003eThe forecasting task involved predicting CO₂ (PM\u003csub\u003e2.5\u003c/sub\u003e) levels 12-time steps ahead (i.e., 60 minutes) using historical data from the preceding \u003cem\u003eh\u003c/em\u003e hours. We evaluated multiple \u003cem\u003eh\u003c/em\u003e configurations, ranging from 1 to 16. To ensure generalizability, model performance was assessed using a leave-one-classroom-out cross-validation strategy. For each iteration, the model was trained on data from 10 classrooms (two nonconsecutive weeks' data series per classroom) and tested on the excluded one. This process was repeated until each classroom's data had served as the test set. All reported metrics, including the coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE), reflect out-of-sample predictions on held-out classrooms. Hyperparameters were optimized using grid search.\u003c/p\u003e\u003cp\u003eIn addition to XGBoost, we tested several alternative regression algorithms, such as support vector regression and random forest. While some models performed adequately under specific configurations, XGBoost consistently outperformed them in terms of predictive accuracy and generalization across classrooms. A transformers-based approach was not considered in our study due to the limited size of our data set, for which XGBoost remains the more practical choice. This modeling strategy aligns with best practices in time series forecasting and environmental data science, where temporal generalization must be rigorously validated.\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e,\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e\u003ch2\u003eCO\u003csub\u003e2\u003c/sub\u003e Concentration Metrics and Airborne Infection Probability\u003c/h2\u003e\u003cp\u003eCO₂-based estimates of airborne infection probability (AIP) provide a practical proxy for ventilation adequacy and airborne transmission risk, particularly when direct airflow data are unavailable. In enclosed classrooms with limited mechanical ventilation, elevated CO₂ generated by exhaled aerosols indicates rebreathed air, potentially containing infectious particles. This method is particularly beneficial in colder seasons, when ventilation is reduced.\u003c/p\u003e\u003cp\u003eRudnick and Milton³⁰ adapted the Wells-Riley equation to use indoor CO₂ as a proxy for rebreathed air. The model estimates the rebreathed fraction from indoor–outdoor CO₂ differences and computes infection probability as:\u003c/p\u003e\u003cdiv id=\"Eque\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Eque\" name=\"EquationSource\"\u003e\n$$\\:P\\:=\\:1\\:-\\:exp\\left(-\\frac{\\stackrel{-}{f}Iq{\\Delta\\:}t}{n}\\right),$$\u003c/div\u003e\u003c/div\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\stackrel{\\prime }{f}\\)\u003c/span\u003e\u003c/span\u003e is the average rebreathed fraction, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:I\\:\\)\u003c/span\u003e\u003c/span\u003ethe number of infectors, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:q\\:\\)\u003c/span\u003e\u003c/span\u003ethe quanta generation rate, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\Delta\\:}t\\)\u003c/span\u003e\u003c/span\u003e the exposure time, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:n\\)\u003c/span\u003e\u003c/span\u003e the number of occupants. See equations 3 and 9 in Rudnick and Milton.\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e We applied this model to real-time CO₂ data from 11 classrooms, assuming that \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:I=1\\)\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:q=1\\:\\)\u003c/span\u003e\u003c/span\u003eand \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:n\\:\\)\u003c/span\u003e\u003c/span\u003eequal to student count. A second scenario included a 37% vaccination rate per school, which is the average rate in SLPMA. Bazant and Bush\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e offer an alternative to compute AIP.\u003c/p\u003e\u003cp\u003eWe also used XGBoost to forecast five minutes ahead using the prior ten minutes of data. This autoregressive model captured temporal risk patterns. To ensure generalizability, we employed leave-one-classroom-out cross-validation. Analysis focused on the first 120 minutes of class (08:00–10:00 h), when infection risk rose notably in both seasons.\u003c/p\u003e\u003ch2\u003eOutdoor PM\u003csub\u003e2.5\u003c/sub\u003e Pollution Sources\u003c/h2\u003e\u003cp\u003eGiven the locations of pollution sources in the SLPMA, we conducted a spatial proximity analysis. We determined the distance between each classroom and nearby pollution sources to estimate potential exposure intensity, considering the proximity of a school to traffic corridors, industrial zones, or biomass burning sites. Even without emission rates, proximity itself is a strong indicator of influence, especially for PM\u003csub\u003e2.5\u003c/sub\u003e, which can vary sharply over short distances. We defined buffer zones around each school, ranging from 100 meters to 3 kilometers, to quantify the number of pollution sources within varying spatial scales. This approach enables the exploration of how local emission pressure may influence IAQ, even in the absence of direct emission measurements. By examining source density across these zones, we can construct a spatial narrative linking environmental context to the observed PM\u003csub\u003e2.5\u003c/sub\u003e data.\u003c/p\u003e\u003cp\u003e \u003cb\u003eAir Sentinel\u003c/b\u003e \u003cb\u003eDeployment and Data Visualization\u003c/b\u003e\u003c/p\u003e\u003cp\u003eA major challenge of our study was implementing an operational IoT network within the SLPMA, where connectivity and networking resources are limited. To address such limited infrastructure, the system was designed as a decentralized, mobile-driven network utilizing IAQ monitors that include calibrated sensors for CO\u003csub\u003e2\u003c/sub\u003e, PM\u003csub\u003e2.5\u003c/sub\u003e, humidity, and temperature. We paired each monitor with a cellphone via Bluetooth. Due to institutional regulations, each classroom was equipped with one monitor and only one cell phone, for which the teacher was responsible. Monitors were positioned strategically to measure breathing-zone air quality (1.0 and 1.5 meters above the floor), avoiding direct interference from pollution sources or factors that could distort measurements. In most cases, monitors were placed on the back wall or near the front wall, close to the teacher’s desk.\u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e The monitor transmits real-time data to the companion mobile app (\u003cem\u003eLinkApp\u003c/em\u003e), which stores, displays, and synchronizes data to the \u003cem\u003eAir Sentinel\u003c/em\u003e cloud platform. The \u003cem\u003eLinkApp\u003c/em\u003e manages data from up to 10 monitors. This lean and highly portable network structure with minimal hardware dependencies may be affected due to the short-range nature of Bluetooth and the variability of Wi-Fi signal strength.\u003c/p\u003e\u003cp\u003eTo support long-term analysis and remote data access, \u003cem\u003eLinkApp\u003c/em\u003e was integrated into the cloud-based \u003cem\u003eAir Sentinel\u003c/em\u003e backend platform that handled secure data ingestion through encrypted protocols and stored sensor readings in a structured database indexed by classroom and timestamp. It also provided alert management, historical trend analysis, and school-wide comparisons across seasons. The backend was designed to be scalable and compliant with data privacy standards.\u003c/p\u003e\u003cp\u003eOn the front end, \u003cem\u003eLinkApp\u003c/em\u003e served as a sensor interface and teacher dashboard, offering immediate feedback on classroom IAQ with visual indicators reflecting safety thresholds (alerts at 1000 ppm and 35 \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\mu\\:\\)\u003c/span\u003e\u003c/span\u003eg/m\u003csup\u003e³\u003c/sup\u003e). Additionally, a web-based \u003cem\u003eAir Sentinel\u003c/em\u003e dashboard allowed school officials and researchers to access graphical summaries, download reports, and analyze exposure patterns over time. This dual-interface design empowered educators to make ventilation decisions during class while also enabling institutional oversight and public health integration. Educators accessed the \u003cem\u003eAir Sentinel\u003c/em\u003e platform through a dedicated dashboard in two progressively enhanced versions; the initial version provided real-time visualization of raw CO₂ and PM\u003csub\u003e2.5\u003c/sub\u003e measurements, along with alert notifications triggered when predefined safety thresholds were exceeded.\u003c/p\u003e\u003cp\u003eThe second \u003cem\u003eAir Sentinel\u003c/em\u003e iteration introduced a more advanced dashboard to analyze processed data and support decision-making. This dashboard allows users to filter by time, window, and classroom, tailoring the analysis to specific contexts. It presents average values for all monitored parameters (CO₂, PM\u003csub\u003e2.5\u003c/sub\u003e, humidity, and temperature) across the selected time and classifies each parameter qualitatively into categories. A dedicated section focuses on CO₂, featuring bar charts that display average CO₂ levels per classroom, enabling comparative analysis and identifying areas that may require improved ventilation. A horizontal summary chart categorizes all measurements into predefined CO₂ quality levels, offering a clear overview of the IAQ situation. A summary panel interprets the data by providing a numerical IAQ quality score and recommendations, such as moderate ventilation, to improve conditions. An explanatory legend connects CO₂ concentration ranges to health and comfort implications, helping stakeholders understand the significance of the observed values. \u003cem\u003eAir Sentinel\u003c/em\u003e supports proactive IAQ management by adapting to diverse environments and data infrastructures, promoting healthier and safer indoor spaces through informed and responsive action.\u003c/p\u003e\u003ch2\u003eCO\u003csub\u003e2\u003c/sub\u003e, PM\u003csub\u003e2.5\u003c/sub\u003e Sensor Calibration and Measurements\u003c/h2\u003e\u003cp\u003eTo ensure the reliability of IAQ measurements, we calibrated both CO\u003csub\u003e2\u003c/sub\u003e and PM\u003csub\u003e2.5\u003c/sub\u003e sensors before their deployment. For CO\u003csub\u003e2\u003c/sub\u003e sensors (range 0–5000 ppm, accuracy ± 50 ppm), which utilize non-dispersive infrared technology, calibration was conducted using a two-point reference procedure. Each device was exposed to a zero-gas environment (high-purity nitrogen) to establish the baseline; this was followed by exposure to a certified gas mixture with a known CO\u003csub\u003e2\u003c/sub\u003e concentration (typically 1000 ppm), allowing for span adjustment. This procedure established a calibration curve for each sensor, which was then validated against ambient outdoor air concentrations assumed to be near 400 ppm. In classrooms where automatic baseline correction algorithms were active, we reviewed sensor placement to ensure periodic exposure to fresh air, avoiding artificially elevated baselines due to continuous occupancy.\u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e For PM\u003csub\u003e2.5\u003c/sub\u003e sensors (range 0-1000 \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\mu\\:\\)\u003c/span\u003e\u003c/span\u003eg/m\u003csup\u003e³\u003c/sup\u003e, accuracy ± 10% of the reading), calibration was conducted via co-location with a reference-grade gravimetric instrument in a controlled indoor setting over a period of two weeks. Sensors recorded optical particles count in parallel with mass concentrations obtained from the reference device. These datasets were analyzed to generate a correction factor using regression modeling, which accounted for temperature and relative humidity, as these factors influence particle scattering and sensor accuracy. Calibration models were embedded in the device firmware before field deployment. Additionally, we performed seasonal calibration validation to account for variations in aerosol composition and ambient moisture, characteristic of the semi-arid conditions in San Luis Potosí.\u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e Temperature and relative humidity sensors were also calibrated using standard techniques.\u003c/p\u003e\u003cp\u003eCO\u003csub\u003e2\u003c/sub\u003e, PM\u003csub\u003e2.5\u003c/sub\u003e, temperature, and relative humidity measurements were continuously recorded over 24-hour periods using monitors equipped with four dedicated sensors corresponding to each environmental parameter. Although data collection spanned the entire day, classroom activities occurred between 08:00 and 14:00 hours, Monday through Friday. Each monitor was connected via Bluetooth or Wi-Fi to the \u003cem\u003eAir Sentinel\u003c/em\u003e platform, which enabled real-time data synchronization, centralized access, and secure storage. Data collection followed institutional standards for privacy protection. Experiments were conducted under the agreement of school authorities and parental consent.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eCompeting interests\u003c/h2\u003e \u003cp\u003eThe authors have no competing interests to declare.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eConceptualization: SR‑C**,** RL‑R and LR‑R. Methodology: SR‑C and RL‑R. Software: SR‑C**,** RL‑R**,** LA**,** CH‑R**,** MAC‑J and JGR‑A. Validation: SR‑C**,** RL‑R**,** LR‑R and CH‑R. Formal analysis: SR‑C**,** RL‑R**,** LR‑R**,** GD**,** KB**,** NG‑H**,** AD**,** LA**,** FM‑C**,** CH‑R and MAC‑J. Investigation: SR‑C**,** RL‑R**,** LR‑R**,** GD**,** KB**,** FM‑C**,** CH‑R and MAC‑J. Resources: SR‑C**,** RL‑R**,** LR‑R**,** NG‑H and CH‑R. Data Curation: SR‑C**,** RL‑R**,** LR‑R**,** NG‑H**,** CH‑R and JGR‑A. Writing ‑ Original Draft: SR‑C**,** RL‑R**,** LR‑R**,** GD**,** KB**,** NG‑H**,** AD**,** LA**,** FM‑C and MAC‑J. Writing ‑ Review \u0026amp; Editing: SR‑C**,** RL‑R**,** LR‑R**,** GD**,** KB**,** NG‑H**,** AD**,** LA**,** FM‑C**,** CH‑R and MAC‑J. Visualization: SR‑C**,** RL‑R**,** CH‑R and MAC‑J. Supervision: SR‑C**,** RL‑R**,** LR‑R and CH‑R. Project administration: SR‑C and RL‑R. Funding: RL‑R.\u003c/p\u003e\u003ch2\u003eAcknowledgements\u003c/h2\u003e \u003cp\u003eWe gratefully acknowledge the advice and support of Dr. Jos\u0026eacute;-Luis Jim\u0026eacute;nez and Dr. Patricia Ripoll of the Aireamos Group, and the funding by generous research grants from the Consejo Potosino de Ciencia y Tecnolog\u0026iacute;a (FME/2021/SO-02/14), and the Balvi Clean Air Initiative (A16).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eCincinelli, A. \u0026amp; Martellini, T. Indoor Air Quality and Health. \u003cem\u003eInt. J. Environ. Res. Public. Health\u003c/em\u003e. \u003cb\u003e14\u003c/b\u003e, 1286 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHonan, D., Gallagher, J., Garvey, J. \u0026amp; Littlewood, J. Indoor Air Quality in Naturally Ventilated Primary Schools: A Systematic Review of the Assessment \u0026amp; Impacts of CO2 Levels. \u003cem\u003eBuildings\u003c/em\u003e \u003cb\u003e14\u003c/b\u003e, 4003 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShukla, A., Indaliya, R. \u0026amp; Tandel, B. N. Effect of Indoor Air Quality on Respiratory Health of Children: An MPPD Model Approach. \u003cem\u003eAerosol Sci. Eng.\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e, 180\u0026ndash;194 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGe, B. et al. Decision tools for schools using continuous indoor air quality monitors: a case study of CO2 in Boston Public Schools. \u003cem\u003eLancet Reg. Health - Am.\u003c/em\u003e \u003cb\u003e48\u003c/b\u003e, 101148 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eP\u0026eacute;rez Velasco, R. \u0026amp; Jarosińska, D. Update of the WHO global air quality guidelines: Systematic reviews \u0026ndash; An introduction. \u003cem\u003eEnviron. Int.\u003c/em\u003e \u003cb\u003e170\u003c/b\u003e, 107556 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSetty, R. et al. Design Guidance for Education Facilities: Prioritization for Advanced Indoor Air Quality Version 2.0. (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUS EPA. The EPA\u0026rsquo;s National School Indoor Air Quality Newsletter - Fall. (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://content.govdelivery.com/accounts/USEPAIAQ/bulletins/3bd7230\u003c/span\u003e\u003cspan address=\"https://content.govdelivery.com/accounts/USEPAIAQ/bulletins/3bd7230\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLv, M., Liu, S., Cao, Q., Zhang, T. \u0026amp; Liu, J. Influence of Ventilation on Indoor Air Quality. in Handbook of Indoor Air Quality (eds Zhang, Y., Hopke, P. K. \u0026amp; Mandin, C.) 1\u0026ndash;38 (Springer Singapore, Singapore, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/978-981-10-5155-5_54-1\u003c/span\u003e\u003cspan address=\"10.1007/978-981-10-5155-5_54-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePeng, Z. \u0026amp; Jimenez, J. L. Exhaled CO\u003csub\u003e2\u003c/sub\u003e as a COVID-19 Infection Risk Proxy for Different Indoor Environments and Activities. \u003cem\u003eEnviron. Sci. Technol. Lett.\u003c/em\u003e \u003cb\u003e8\u003c/b\u003e, 392\u0026ndash;397 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCDC. Ventilation Can Reduce Exposure to Respiratory Viruses in Indoor Spaces. \u003cem\u003eNational Center for Immunization and Respiratory Diseases\u003c/em\u003e (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.cdc.gov/ncird/whats-new/ventilation-respiratory-viruses.html\u003c/span\u003e\u003cspan address=\"https://www.cdc.gov/ncird/whats-new/ventilation-respiratory-viruses.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZaniboni, L. \u0026amp; Albatici, R. Natural and Mechanical Ventilation Concepts for Indoor Comfort and Well-Being with a Sustainable Design Perspective: A Systematic Review. \u003cem\u003eBuildings\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 1983 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMustafa, M., Cook, M. J. \u0026amp; McLeod, R. S. A critical review of ventilation effectiveness in naturally ventilated spaces from the perspective of sustainability and health. \u003cem\u003eBuild. Environ.\u003c/em\u003e \u003cb\u003e270\u003c/b\u003e, 112471 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBecerra, J. A. et al. Identification of potential indoor air pollutants in schools. \u003cem\u003eJ. Clean. Prod.\u003c/em\u003e \u003cb\u003e242\u003c/b\u003e, 118420 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVargas-Legarda, Y. A., Toro-Mart\u0026iacute;nez, A. K. \u0026amp; Rojas-Roa, N. Y. Fajardo-Monta\u0026ntilde;a, O. A. School Children\u0026rsquo;s exposure to 〖PM〗_(2.5) in a high pollution area of Bogot\u0026aacute;, Colombia. \u003cem\u003eIng. E Investig\u003c/em\u003e. \u003cb\u003e43\u003c/b\u003e, e96125 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMart\u0026iacute;nez, L. et al. Indoor-outdoor concentrations of fine particulate matter in school building microenvironments near a mine tailing deposit. \u003cem\u003eAIMS Environ. Sci.\u003c/em\u003e \u003cb\u003e3\u003c/b\u003e, 752\u0026ndash;764 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDiaz, M., Cools, M., Trebilcock, M., Piderit-Moreno, B. \u0026amp; Attia, S. Effects of Climatic Conditions, Season and Environmental Factors on CO2 Concentrations in Naturally Ventilated Primary Schools in Chile. \u003cem\u003eSustainability\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 4139 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSahin, C., Dogan, R., Yildiz, T. \u0026amp; Sofuoglu, S. C. M. Indoor environmental quality in naturally ventilated schools of a dusty region: Excess health risks and effect of heating and desert dust transport. \u003cem\u003eIndoor Air\u003c/em\u003e \u003cb\u003e32\u003c/b\u003e, (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGil-Baez, M. et al. Natural ventilation in classrooms for healthy schools in the COVID era in Mediterranean climate. \u003cem\u003eBuild. Environ.\u003c/em\u003e \u003cb\u003e206\u003c/b\u003e, 108345 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOertel, M. et al. Drought Propagation in Semi-Arid River Basins in Latin America: Lessons from Mexico to the Southern Cone. \u003cem\u003eWater\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 1564 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRose, O. G., Bousiotis, D., Rathbone, C. \u0026amp; Pope, F. D. Investigating Indoor Air Pollution Sources and Student\u0026rsquo;s Exposure Within School Classrooms: Using a Low-Cost Sensor and Source Apportionment Approach. \u003cem\u003eIndoor Air\u003c/em\u003e 5544298 (2024). (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHiggins, C., Kumar, P. \u0026amp; Morawska, L. Indoor air quality monitoring and source apportionment using low-cost sensors. \u003cem\u003eEnviron. Res. Commun.\u003c/em\u003e \u003cb\u003e6\u003c/b\u003e, 012001 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eApostolopoulos, I. D. et al. Monitoring of Indoor Air Quality in a Classroom Combining a Low-Cost Sensor System and Machine Learning. \u003cem\u003eChemosensors\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 148 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eINEGI. Directorio Nacional de Unidades Econ\u0026oacute;micas. DENUE. \u003cem\u003eCensos Econ\u0026oacute;micos 2014\u003c/em\u003e (2014). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.inegi.org.mx/app/mapa/denue/default.aspx\u003c/span\u003e\u003cspan address=\"https://www.inegi.org.mx/app/mapa/denue/default.aspx\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAllen, J., Spengler, J., Jones, E. \u0026amp; Cedeno-Laurent, J. 5-step guide to checking ventilation rates in classrooms. (2020). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://schools.forhealth.org/wp-content/uploads/sites/19/2020/10/Harvard-Healthy-Buildings-program-How-to-assess-classroom-ventilation-10-30-2020.pdf-EN.pdf\u003c/span\u003e\u003cspan address=\"https://schools.forhealth.org/wp-content/uploads/sites/19/2020/10/Harvard-Healthy-Buildings-program-How-to-assess-classroom-ventilation-10-30-2020.pdf-EN.pdf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAllen, J. G. et al. Associations of Cognitive Function Scores with Carbon Dioxide, Ventilation, and Volatile Organic Compound Exposures in Office Workers: A Controlled Exposure Study of Green and Conventional Office Environments. \u003cem\u003eEnviron. Health Perspect.\u003c/em\u003e \u003cb\u003e124\u003c/b\u003e, 805\u0026ndash;812 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, T., Guestrin, C. \u0026amp; XGBoost: A Scalable Tree Boosting System. in \u003cem\u003eProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining\u003c/em\u003e 785\u0026ndash;794ACM, San Francisco California USA, (2016). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1145/2939672.2939785\u003c/span\u003e\u003cspan address=\"10.1145/2939672.2939785\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eG\u0026eacute;ron, A. \u003cem\u003eHands-on Machine Learning with Scikit-Learn, Keras, and TensorFlow\u003c/em\u003e. (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKuhn, M. \u0026amp; Johnson, K. \u003cem\u003eApplied Predictive Modeling\u003c/em\u003e (Springer New York, 2013). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/978-1-4614-6849-3\u003c/span\u003e\u003cspan address=\"10.1007/978-1-4614-6849-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHyndman, R. J. \u0026amp; Athanasoupulos, G. \u003cem\u003eForecasting: Principles and Practice (3rd Ed)\u003c/em\u003eOTexts, Melbourne, Australia,. (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRudnick, S. N. \u0026amp; Milton, D. K. Risk of indoor airborne infection transmission estimated from carbon dioxide concentrationAbstract. \u003cem\u003eIndoor Air\u003c/em\u003e. \u003cb\u003e13\u003c/b\u003e, 237\u0026ndash;245 (2003).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBazant, M. Z. \u0026amp; Bush, J. W. M. A guideline to limit indoor airborne transmission of COVID-19. \u003cem\u003eProc. Natl. Acad. Sci.\u003c/em\u003e 118, e2018995118 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, M. et al. The effect of using phase change materials in solar panel cooling to provide green and sustainable energy of a building. \u003cem\u003eJ. Build. Eng.\u003c/em\u003e \u003cb\u003e57\u003c/b\u003e, 104933 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCai, Q. et al. Evaluation of Low-Cost CO2 Sensors Using Reference Instruments and Standard Gases for Indoor Use. \u003cem\u003eSensors\u003c/em\u003e \u003cb\u003e24\u003c/b\u003e, 2680 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePark, D., Yoo, G. W., Park, S. H. \u0026amp; Lee, J. H. Assessment and Calibration of a Low-Cost PM2.5 Sensor Using Machine Learning (HybridLSTM Neural Network): Feasibility Study to Build an Air Quality Monitoring System. \u003cem\u003eAtmosphere\u003c/em\u003e 12, 1306 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSadrizadeh, S. et al. Indoor air quality and health in schools: A critical review for developing the roadmap for the future school environment. \u003cem\u003eJ. Build. Eng.\u003c/em\u003e \u003cb\u003e57\u003c/b\u003e, 104908 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbdullah, H. K. \u0026amp; Alibaba, H. Z. A Performance-Based Window Design and Evaluation Model for Naturally Ventilated Offices. \u003cem\u003eBuildings\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 1141 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRagab, A., Hassieb, M. M. \u0026amp; Mohamed, A. F. Exploring the impact of window design and ventilation strategies on air quality and thermal comfort in arid educational buildings. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e15\u003c/b\u003e, 19596 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGarcia, M. S., Souza, R. V. G. \u0026amp; Silva, F. J. Towards Health and Comfort: Architectural Design and Indoor Air Quality in Naturally Ventilated Classrooms. in \u003cem\u003eProceedings of CIRMARE 2023\u003c/em\u003e (eds. Lanzinha, J. C. G. \u0026amp; Qualharini, E. L.) vol. 444 403\u0026ndash;413Springer Nature Switzerland, Cham, (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLawal, A. A., Dare-Abel, O. A. \u0026amp; Ajayi, O. O. A Review of Indoor Environmental Quality (Ieq) Parameters and Their Impact on Occupant Satisfaction in Educational Buildings. \u003cb\u003e9\u003c/b\u003e, (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFisk, W. J. The ventilation problem in schools: literature review. \u003cem\u003eIndoor Air\u003c/em\u003e. \u003cb\u003e27\u003c/b\u003e, 1039\u0026ndash;1051 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSettimo, G. et al. CO2 Levels in Classrooms: What Actions to Take to Improve the Quality of Environments and Spaces. \u003cem\u003eSustainability\u003c/em\u003e \u003cb\u003e16\u003c/b\u003e, 8619 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBadyda, A. \u0026amp; Rogulski, M. Closing Access Streets to Schools for Vehicular Traffic\u0026mdash;Does. \u003cem\u003eIt Affect. Air Quality? Sustainability\u003c/em\u003e. \u003cb\u003e16\u003c/b\u003e, 3380 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlameddine, I., Gebrael, K., Hanna, F. \u0026amp; El-Fadel, M. Quantifying indoor PM2.5 levels and its sources in schools: What role does location, chalk use, and socioeconomic equity play ? \u003cem\u003eAtmospheric Pollut Res.\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 101375 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlves, C., Nunes, T., Silva, J. \u0026amp; Duarte, M. Comfort Parameters and Particulate Matter (PM10 and PM2.5) in School Classrooms and Outdoor Air. \u003cem\u003eAerosol Air Qual. Res.\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 1521\u0026ndash;1535 (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePan, H. \u0026amp; Wu, C. Bayesian optimization\u0026thinsp;+\u0026thinsp;XGBoost based life cycle carbon emission prediction for residential buildings\u0026mdash;An example from Chengdu, China. \u003cem\u003eBuild. Simul.\u003c/em\u003e \u003cb\u003e16\u003c/b\u003e, 1451\u0026ndash;1466 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, R., Li, Y., Zhang, A. L., Wang, Y. \u0026amp; Molina, M. J. Identifying airborne transmission as the dominant route for the spread of COVID-19. \u003cem\u003eProc. Natl. Acad. Sci.\u003c/em\u003e 117, 14857\u0026ndash;14863 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShah, K. B. et al. Evaluating Indoor Air Quality in Residential Environments: A Study of PM2.5 and CO2 Dynamics Using Low-Cost Sensors. \u003cem\u003eEnvironments\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e, 237 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDai, Z., Yuan, Y., Zhu, X. \u0026amp; Zhao, L. A Method for Predicting Indoor CO2 Concentration in University Classrooms: An RF-TPE-LSTM Approach. \u003cem\u003eAppl. Sci.\u003c/em\u003e \u003cb\u003e14\u003c/b\u003e, 6188 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDai, Y. et al. Synergistic effects of CO2 and air pollutants from ship emissions in Shanghai, China: Spatial-temporal characteristics, prediction assessment, policy implications. \u003cem\u003eJ. Environ. Manage.\u003c/em\u003e \u003cb\u003e376\u003c/b\u003e, 124417 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu, W. et al. Indoor PM2.5 forecasting and the association with outdoor air pollution: a modelling study based on sensor data in Australia. \u003cem\u003ePreprint at.\u003c/em\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.48550/arXiv.2405.07404\u003c/span\u003e\u003cspan address=\"10.48550/arXiv.2405.07404\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eElbayoumi, M. \u0026amp; Ramli, N. A. Fitri Md Yusof, N. F. Development and comparison of regression models and feedforward backpropagation neural network models to predict seasonal indoor PM2.5\u0026ndash;10 and PM2.5 concentrations in naturally ventilated schools. \u003cem\u003eAtmospheric Pollut Res.\u003c/em\u003e \u003cb\u003e6\u003c/b\u003e, 1013\u0026ndash;1023 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWHO. \u003cem\u003eWHO Global Air Quality Guidelines: Particulate Matter (PM2.5 and PM10), Ozone, Nitrogen Dioxide, Sulfur Dioxide and Carbon Monoxide\u003c/em\u003e (World Health Organization, 2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChaudhry, S. K. et al. Influence of seasonal variation on spatial distribution of PM2.5 concentration using low-cost sensors. \u003cem\u003eEnviron. Monit. Assess.\u003c/em\u003e \u003cb\u003e196\u003c/b\u003e, 1234 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eElbayoumi, M. et al. Multivariate methods for indoor PM10 and PM2.5 modelling in naturally ventilated schools buildings. \u003cem\u003eAtmos. Environ.\u003c/em\u003e \u003cb\u003e94\u003c/b\u003e, 11\u0026ndash;21 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBuonanno, G., Stabile, L. \u0026amp; Morawska, L. Estimation of airborne viral emission: Quanta emission rate of SARS-CoV-2 for infection risk assessment. \u003cem\u003eEnviron. Int.\u003c/em\u003e \u003cb\u003e141\u003c/b\u003e, 105794 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVouriot, C. V. M., Burridge, H. C., Noakes, C. J. \u0026amp; Linden, P. F. Seasonal variation in airborne infection risk in schools due to changes in ventilation inferred from monitored carbon dioxide. \u003cem\u003eIndoor Air\u003c/em\u003e. \u003cb\u003e31\u003c/b\u003e, 1154\u0026ndash;1163 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMaantay, J. Mapping environmental injustices: pitfalls and potential of geographic information systems in assessing environmental health and equity. \u003cem\u003eEnviron. Health Perspect.\u003c/em\u003e \u003cb\u003e110\u003c/b\u003e, 161\u0026ndash;171 (2002).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChakraborty, J. \u0026amp; Maantay, J. A. Proximity Analysis for Exposure Assessment in Environmental Health Justice Research. in Geospatial Analysis of Environmental Health (eds Maantay, J. A. \u0026amp; McLafferty, S.) 111\u0026ndash;138 (Springer Netherlands, Dordrecht, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/978-94-007-0329-2_5\u003c/span\u003e\u003cspan address=\"10.1007/978-94-007-0329-2_5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. (2011).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUS EPA, O. Reference Guide for Indoor Air Quality in Schools. (2014). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.epa.gov/iaq-schools/reference-guide-indoor-air-quality-schools\u003c/span\u003e\u003cspan address=\"https://www.epa.gov/iaq-schools/reference-guide-indoor-air-quality-schools\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMatthaios, V. N. et al. The effects of urban green space and road proximity to indoor traffic-related PM2.5, NO2, and BC exposure in inner-city schools. \u003cem\u003eJ. Expo Sci. Environ. Epidemiol.\u003c/em\u003e \u003cb\u003e34\u003c/b\u003e, 745\u0026ndash;752 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eINECC. \u003cem\u003eEvaluaci\u0026oacute;n de Contaminantes Atmosf\u0026eacute;ricos en San Luis Potos\u0026iacute;\u003c/em\u003e. 32 (2018). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.gob.mx/cms/uploads/attachment/file/399489/Informe_Ejecutivo_SLP_Rev3__1_.pdf\u003c/span\u003e\u003cspan address=\"https://www.gob.mx/cms/uploads/attachment/file/399489/Informe_Ejecutivo_SLP_Rev3__1_.pdf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAl Osman, M., Yang, F. \u0026amp; Massey, I. Y. Exposure routes and health effects of heavy metals on children. \u003cem\u003eBioMetals\u003c/em\u003e \u003cb\u003e32\u003c/b\u003e, 563\u0026ndash;573 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBora, J. et al. IGI Global Scientific Publishing,. Health Effects of Heavy Metals Contamination in Children. in \u003cem\u003eNanotechnology Applications and Innovations for Improved Soil Health\u003c/em\u003e 254\u0026ndash;275 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.4018/979-8-3693-1471-5.ch012\u003c/span\u003e\u003cspan address=\"10.4018/979-8-3693-1471-5.ch012\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGartland, N. et al. The Effects of Traffic Air Pollution in and around Schools on Executive Function and Academic Performance in Children: A Rapid Review. \u003cem\u003eInt. J. Environ. Res. Public. Health\u003c/em\u003e. \u003cb\u003e19\u003c/b\u003e, 749 (2022).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTable 1. Ventilation features of classrooms and other rooms\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"8\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eClassrooms (N = 73)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eFeatures\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eStd\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eMin\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e25%\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e50%\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e75%\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eMax\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMin ACH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9.20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e11.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e18.90\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMax ACH (m)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e14.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e10.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e14.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e18.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e29.28\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAverage ACH (m)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e22.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e11.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e13.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e18.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e34.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e40.20\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eScore\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2.40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e12.12\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"8\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eOther rooms (N = 33)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eFeatures\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eStd\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eMin\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e25%\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e50%\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e75%\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eMax\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMin ACH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.80\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6.10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e8.60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e16.80\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMax ACH (m)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e11.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6.53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7.20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9.42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e15.36\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e29.52\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAverage ACH (m)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e17.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e11.65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e12.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e28.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e41.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eScore\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.87\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eKey: Air changes per hour (ACH). A score lower than one indicates \u003cem\u003epoor ventilation\u003c/em\u003e. A score between one and two indicates \u003cem\u003eacceptable ventilation\u003c/em\u003e. A score greater than two indicates \u003cem\u003eexcellent ventilation\u003c/em\u003e.\u003c/p\u003e\n"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Indoor air quality, Carbon dioxide forecasting, Classroom ventilation, PM2.5 exposure, Machine learning regression, Airborne infection probability, IoT environmental monitoring","lastPublishedDoi":"10.21203/rs.3.rs-8379552/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8379552/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIndoor air quality (IAQ) is neglected in schools of the Global South. Carbon dioxide (CO\u003csub\u003e2\u003c/sub\u003e) and particulate matter (PM\u003csub\u003e2.5\u003c/sub\u003e) are indicators of ventilation and exposure risk to airborne infections, respectively. We deployed \u003cem\u003eAir Sentinel\u003c/em\u003e, a decentralized mobile-driven IoT network, to monitor these indicators across 106 spaces in 53 elementary schools in the San Luis Potos\u0026iacute; Metropolitan Area, Mexico. Machine learning was used to forecast CO\u003csub\u003e2\u003c/sub\u003e and PM\u003csub\u003e2.5\u003c/sub\u003e concentrations one hour in advance, and the Wells-Riley model, based on CO\u003csub\u003e2\u003c/sub\u003e concentration, to estimate airborne infection probability (AIP). IAQ was poor, acceptable, and excellent in 27.3%, 36.4%, and 67.2% of the classrooms, respectively. During the cold season, 97.67% of the classroom CO\u003csub\u003e2\u003c/sub\u003e levels were typical (400\u0026ndash;1000 ppm); in the hot season, 67.67% of the classroom CO\u003csub\u003e2\u003c/sub\u003e levels were typical, and 1.19% exceeded the high exposure threshold (\u0026gt;\u0026thinsp;2000 ppm). Classroom CO\u003csub\u003e2\u003c/sub\u003e dynamics exhibited low temporal synchrony. The strongest forecasting performance for CO\u003csub\u003e2\u003c/sub\u003e occurred in the hot season, but the PM\u003csub\u003e2.5\u003c/sub\u003e forecast failed in either season while AIP increased during the first two hours of class in both seasons. The successful CO\u003csub\u003e2\u003c/sub\u003e forecasting model has potential for real-time IAQ management in the cold season. The failure to forecast PM\u003csub\u003e2.5\u003c/sub\u003e levels suggests that localized sources drive their dynamics. We conclude that the \u003cem\u003eAir Sentinel\u003c/em\u003e network is a convenient classroom IAQ monitor in the Global South.\u003c/p\u003e","manuscriptTitle":"Air Sentinel: An IoT-Based Platform for Monitoring Indoor Air Quality in Elementary Schools of the Global South","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-23 15:31:08","doi":"10.21203/rs.3.rs-8379552/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-12-23T10:27:29+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-12-20T08:37:02+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-12-20T08:36:14+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-12-16T19:59:25+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"1f353a1a-8bc7-461e-9de8-8e790e72bb73","owner":[],"postedDate":"December 23rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":59913838,"name":"Earth and environmental sciences/Climate sciences"},{"id":59913839,"name":"Earth and environmental sciences/Environmental sciences"},{"id":59913840,"name":"Earth and environmental sciences/Environmental social sciences"}],"tags":[],"updatedAt":"2026-03-31T14:53:28+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-23 15:31:08","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8379552","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8379552","identity":"rs-8379552","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}