{"paper_id":"4ada67ca-ef7c-4406-ba83-d184cb4d0703","body_text":"Prophet vs. ARIMA for Forecasting Student Performance from Social-Media Usage | 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 Research Article Prophet vs. ARIMA for Forecasting Student Performance from Social-Media Usage Michael Adelani Adewusi, ZAHARAH NABUNYA This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7879434/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract We evaluate time-series forecasting approaches for predicting end-of-semester examination performance using synthetic data to dynamically measure social-media usage among undergraduate students at an institution in Uganda. Building on an additive decomposition framework, Facebook’s Prophet is specified with trend, weekly/semester seasonality, and holiday regressors reflecting national holidays and university breaks. Benchmarks include ARIMA. Model is trained with rolling cross-validation; accuracy is assessed via MAE, RMSE, and MAPE. Additionally disaggregating usage by purpose (academic vs. recreational) to test for non-linear effects and timing asymmetries around high-stakes periods. Results indicate that Prophet consistently outperforms ARIMA, particularly near changepoints (pre-exam spikes) and during holiday transitions. Purpose-specific features improve forecast accuracy and reveal thresholds were recreational use shifts from neutral to detrimental. The study further demonstrates a lightweight early-warning protocol that converts forecasts into actionable advisories for academic support units. The study advances methodological practice in Sub-Saharan higher education by moving beyond cross-sectional correlations to dynamic forecasting and offers a reproducible Python pipeline institutions to adapt for local quality-assurance and student-success initiatives. Prophet model ARIMA Social media usage Academic performance Rolling cross-validation Higher education in Uganda Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Introduction Across all universities, social media is deeply woven into students’ study habits, peer support, and everyday life. The Platforms enable rapid access to explanations, collaborative note-taking, and instructor–student micro-communities but they also compete with focused study time and sleep, particularly during high-stakes windows such as revision weeks and examinations (Adewusi et al., 2025 ). Empirical findings are mixed in that some studies link social networking to lower grades, while others show learning benefits when usage is structured for pedagogy. This heterogeneity suggests that static, cross-sectional snapshots of “hours online vs. GPA” are too blunt to capture patterns that change week-to-week and surge around assessments (Kirschner & Karpinski, 2010 ; Junco et al., 2011 ). Institutions need methods that treat behaviour and performance as dynamic processes and that generate forward-looking signals administrators can act on. Recent reviews and meta-analyses underline precisely this complexity, reporting negative associations for problematic or high-frequency use alongside positive effects where social media is embedded in learning design. (Kirschner & Karpinski, 2010 ; Junco et al., 2011 ; Masalimova et al., 2023 ; Adewusi et al., 2024 ; Kuş et al., 2025 ; Salari et al., 2025 ; Hsieh et al., 2025 ). As the sub-Saharan Africa continues to experience steady growth in smartphone adoption and mobile internet access, with most learners engaging via 3G/4G and the 5G availability in urban centres, this context amplifies both the opportunity and the risk in which collaborative learning affordances are expanding, yet attention-splitting entertainment content is increasingly ubiquitous that often on the same device and within the same apps students use for study-related activity (GSMA, 2024 ; Adewusi et al., 2022 ). Planning effective student-success interventions therefore benefits from forecasts that anticipate peaks and troughs in risk aligned to the academic calendar. (GSMA, 2024 ). Time-series forecasting provides a principled alternative to cross-sectional association studies by learning structure from historical dynamics and projecting them forward (Taylor & Letham, 2017). For academic performance forecasting conditioned on social-media behaviour, three features are crucial. First, flexible trend modelling is needed to accommodate structural breaks in such that as sudden behavioural shifts before exams or changes in platform algorithms. Second, seasonality must be modelled at multiple frequencies (e.g., weekly rhythms tied to class schedules and semester-level cycles). Third, holiday and break effects should be represented explicitly, as these can induce sharp deviations in both media use and study intensity (Hyndman & Athanasopoulos, 2021 ; Adewusi et al., 2021 ). Prophet, a decomposable additive model with piecewise trends, automatic changepoint detection, multi-scale seasonality, and user-defined holiday regressors was designed for these requirements. Classical ARIMA remain strong baselines for stationary or near-stationary series with stable autocorrelation structures and thus serve as critical comparators for assessing whether Prophet’s additional flexibility yields tangible forecasting gains in this educational context. (Taylor & Letham, 2017; Hyndman & Athanasopoulos, 2021 ). Rather than a single train–test split, rolling-origin also called rolling cross-validation, refits models as new observations arrive and scores forecasts at relevant horizons which is a closer analogue to how an institution would monitor risk week-by-week. This approach offers more stable accuracy estimates and highlights where models struggle (e.g., near changepoints). Accuracy metrics should balance interpretability (e.g., MAPE) with sensitivity to large errors (e.g., RMSE) and absolute scale (MAE). (Tashman, 2000 ; Hyndman & Athanasopoulos, 2021 ). High-frequency traces of student behaviour can be sensitive, dispersed across private devices and third-party platforms, and governed by evolving regulatory expectations. Synthetic data has emerged as a pragmatic pathway to advance methods without exposing personal data (Lyandres, 2022; NIST, 2018 ). It can mimic the statistical structure of real datasets while reducing re-identification risk, particularly when combined with privacy-enhancing technologies (PETs) and differential privacy. Recent guidance and scholarship in education and the public sector highlight synthetic data’s role in enabling analysis and sharing under robust safeguards and a “data protection by design” posture (Lyandres, 2022; NIST, 2018 ). In this study, we therefore evaluate forecasting methods on synthetic series that encode realistic weekly rhythms, semester structure, and holiday shocks for cohorts, along with purpose-specific features (academic vs. recreational use). This design permits transparent benchmarking, stress-testing under known “ground-truth” regimes (e.g., non-linear overload effects), and full reproducibility for institutions seeking to replicate the pipeline with their own ethically sourced aggregates. (Lyandres, 2022; NIST, 2018 ; Jisc, 2025). Research Question - Does the Prophet outperform ARIMA in forecasting academic performance? - How does the nature of social media affect the predictions? Methods The study’s dataset was developed as a reproducible Python pipeline that mirrors an institutional analytics workflow, from design to deployment. A specified academic calendar (four 17-week semesters plus recess) and a built holiday/recess table, then generated weekly, cohort-level synthetic series with three usage signals (minutes, session frequency, academic/recreational mix) and a performance proxy driven by a distributed lag of a latent “study-time efficiency” term. The generator introduced semester and weekly seasonality, revision/exam boosts, recess/holiday dips, and a quadratic overload penalty for excess recreational use, all under fixed random seeds for reproducibility. The implementation of two model families, the Prophet (trend changepoints, weekly and semester Fourier seasonality, holiday regressors, and extra regressors for academic/recreational use) and ARIMAX baselines inside an expanding, rolling-origin cross-validation loop that refit weekly and produced 1–4-week-ahead forecasts. Accuracy was summarized with MAE, RMSE, and MAPE on exam-week targets, the sensitivity checks varied Prophet changepoint priors/Fourier orders and ARIMAX seasonal periods. A lightweight “early-warning” layer mapped forecasts to risk flags with simple, and purpose-aware advisories. The codebase (pandas, NumPy, statsmodels, Prophet) was version-controlled, documented, and emits CSV artefacts and figures, enabling end-to-end reproducibility while keeping data aggregate-only for privacy and ethical compliance. Study Design and Setting The study conducted a methodological evaluation of time-series forecasting approaches for predicting end-of-semester examination performance using synthetic, using python application to cohort-level data reflecting the rhythms of undergraduate study at an institution in Uganda. The focus is on weekly dynamics over four contiguous semesters, allowing models to learn (and exploit) multiple seasonality (weekly, semester-level), structural breaks (e.g., pre-exam shifts), and holiday effects typical of the Ugandan academic calendar. All analyses were performed at the cohort week level (no individual-level records), enabling full reproducibility and privacy preservation while emulating realistic educational signals. Data and Measures The python code generated weekly observations (t = 1,…,T, T ≈ 96) with the following constructs: Social-media usage Frequency: average daily sessions per student (freq t ​). Duration: average daily minutes (mins t ​). Purpose mix: proportion of usage devoted to academic activities vs. recreational content. We denote acad_share t ∈ [0,1] and recr_share t = 1 − acad_share t ​. Purpose-specific minutes are acad_mins t = mins t × acad_share t and recr_mins t = mins t × (1 − acad_share t ). Academic outcome Exam performance (score t ​) which is a standardized weekly performance proxy that is used for forecasting throughout and evaluated only on exam-week targets (mid-semester tests and finals). The series is constructed from a distributed lag of study-time efficiency over the preceding 1–3 weeks and an “exam-week” indicator. Synthetic series embed: (i) weekly seasonality; (ii) semester structure (steady weeks to revision ramp-up to exam weeks to recess); (iii) holiday shocks; and (iv) non-linear overload for recreational use beyond a threshold τ. The latent efficiency term increases with academic use but incurs a quadratic penalty when recr_mins t > τ, creating realistic changepoints before examinations. Holiday Calendar We constructed a Prophet-compatible holiday table including Ugandan fixed-date holidays (e.g., New Year’s Day, NRM Liberation Day, International Women’s Day, Labour Day, Uganda Martyrs’ Day, National Heroes’ Day, Independence Day, Christmas, Boxing Day) and movable feasts (Good Friday, Easter Monday, Eid al-Fitr, Eid al-Adha; dated for the study window). University recess periods between semesters are encoded as holiday-like spans. A weekly holiday_flag equals 1 if any holiday occurs within that week (Monday-anchored). Model Specifications Prophet (primary model) Prophet’s additive decomposition was adopted with the following configuration: Trend: piecewise linear with automatic changepoint detection (default prior; tuned in sensitivity analyses). Seasonality: built-in weekly seasonality; custom semester seasonality via Fourier terms with period equal to semester length (e.g., 17 weeks; typical harmonics K = 8). Yearly seasonality is disabled given the horizon. Holidays: national holidays and recess intervals are entered as event regressors, with ± 1-week windows to capture anticipation/lag effects. Extra regressors: standardized acad_log and recr_log as additive terms with modest prior scales to regularize their influence. Uncertainty: parameter uncertainty is enabled for interval forecasts. ARIMA (benchmarks) ARIMA via a compact box–Jenkins workflow was estimated: Stationarity and differencing: consider d ∈ {0,1} and a semester pseudo-seasonal difference D ∈ {0,1} with seasonal period s = 17. Order selection: grid over (p,q,P,Q) ∈ {0,1,2} using AICc and residual diagnostics (Ljung–Box). Exogenous regressors: holiday dummies and standardized acad_log/recr_log are included as X in ARIMAX variant. Diagnostics: inspect ACF/PACF of residuals; verify approximate homoscedasticity. The default ARIMAX benchmark used in the appendix is (1,1,1) (1,0,1) 17 sensitivity grids are straightforward. Model Training and Rolling Cross-Validation Initial window: the first 40 weeks to fit initial models. Rolling scheme: the origin advances by one week; at each origin models are refit on all past data and produce forecasts for horizons h ∈ {1,2,3,4} weeks ahead (supporting a one-month early-warning window). Targets: metrics are computed only on exam-week forecasts, reflecting decision-relevant outcomes. Leakage control: all scalers, transforms, and hyperparameters are recomputed using training-fold only information at each origin. For Prophet, future exogenous regressors (acad_log, recr_log) are operationalized via a simple hold-forward of the most recent two-week average; institutions can replace this with scenario inputs or short-horizon forecasts of usage. Performance Metrics Three complementary measures: Early-Warning Operationalization At each origin is map by the h-week-ahead forecast of score to a risk probability of falling below a policy threshold (e.g., 0.5 SD below the historical mean). Cohorts are flagged when exceedance probability > 0.60. Alerts are accompanied by feature attributions (comparative influence of academic vs. recreational use), providing actionable guidance to academic support units for targeted nudges in the two to four weeks preceding assessments. All analyses are implemented in Python using Prophet for decomposable forecasting and statsmodels for ARIMAX. The full code used to generate data, construct the holiday calendar, and run rolling evaluation can be provided. The script emits three artefacts for direct inclusion in the Results section: rolling_metrics_per_h.csv (per-horizon MAE/RMSE/MAPE), rolling_metrics_overall.csv (pooled metrics), and rolling_point_forecasts.csv (per-origin/horizon forecast records) Results Table 1 Variables, definitions, summary stats. Variable Definition Mean SD Min Max mins Avg. daily minutes on social media 90.43 43.52 14.85 160.05 freq Avg. daily sessions 9.08 3.12 3.70 15.08 acad_share Proportion of use for academic purposes 0.46 0.06 0.31 0.59 acad_mins Academic minutes = mins X acad_share 41.53 20.14 7.16 78.44 recr_mins Recreational minutes = mins X (1 - acad_share) 48.90 24.26 7.69 90.51 score Weekly performance proxy (0–100); evaluated on exam weeks 70.09 3.07 63.49 76.38 holiday_flag 1 if any national holiday in week 0.00 0.00 0.00 0.00 revision_week 1 in revision weeks 0.11 0.31 FALSE TRUE exam_week 1 in exam weeks 0.11 0.31 FALSE TRUE recess_week 1 in recess weeks 0.11 0.31 FALSE TRUE Table 2 Cross-validated accuracy (Prophet vs. ARIMA) Horizon Model MAE RMSE MAPE(%) 1 Prophet 2.4 3 3.1 1 ARIMAX 3.2 4.1 4.5 2 Prophet 2.7 3.3 3.6 2 ARIMAX 3.6 4.6 5 3 Prophet 3 3.7 4 3 ARIMAX 4 5.1 5.7 4 Prophet 3.3 4.1 4.4 4 ARIMAX 4.3 5.5 6.1 Overall Prophet 2.9 3.5 3.8 Overall ARIMAX 3.8 4.8 5.3 Discussion Our problem combines changepoints, multi-scale seasonality, and calendar shocks in which a pattern Prophet was designed to handle. Prophet models trend as piecewise linear segments and automatically place the changepoints where the series bends, such as the revision ramp and the pre-exam surge in which it reduces bias exactly when autoregressive models tend to mis-specify the data-generating process and this was in agreement to Taylor and Letham, (2017). However, in contrast, ARIMA relies on a relatively stable autocorrelation structure in such that when behavior pivots quickly in the fortnight before exams, that assumption is often violated, inflating forecast error during analysis. The Prophet’s ability to represent multiple seasonality concurrently in having weekly cycles layered over a semester arc through Fourier terms without fragile seasonal differencing or complicated seasonal orders are great welcoming development and this was in consonant with Taylor and Letham, (2017), and Hyndman and Athanasopoulos, ( 2021 ) studies. This aligns with best practice to validate with rolling-origin cross-validation, which checks that patterns survive refits over time in the study (Hyndman & Athanasopoulos, 2021 ). From the study, Prophet treats holidays and recess periods as explicit regressors with user-defined windows. This transparency is crucial in our contextual wise where public holidays and university breaks in Uganda introduce short, irregular shocks that should not be “absorbed” into generic autoregressive persistence as support by Taylor and Letham, (2017) findings. By adding purpose-specific regressors, a standardized logs of academic and recreational minutes will lets Prophet act like a regularized, additive model in which interpretable covariates complement calendar structures. There is no not claim that Prophet is universally superior but large-scale forecast competitions repeatedly show no single method dominates in Prophet, which is combinations and exponential-smoothing but ARIMA families remain formidable, especially on shorter, stationary series with a single dominant seasonal frequency (Makridakis, Spiliotis, & Assimakopoulos, 2018 ; Makridakis et al., 2022 ). This is precisely why we benchmarked against ARIMA and reported uncertainty, that is the method choice should be problem-fit with cross-validated evidence, not brand loyalty (Hyndman & Athanasopoulos, 2021 ). The academic decision window is two to four weeks before examinations that is enough time to schedule revision clinics, organize peer-led sessions, or coordinate advisor outreach as supported by Hyndman an Athanasopoulos, )2021). In our rolling-origin evaluation, Prophet’s error advantage is largest at horizons h = 2–4, which translates into tighter risk estimates exactly when action is feasible (Hyndman & Athanasopoulos, 2021 ). Rather than moralizing “reduce social media,” our results support a practical nudge students recognize as doable for which in the last two weeks before exams, the students keep recreational scrolling near or below approximately 1 hour per day and redirect 60–90 minutes toward academic use like having peer explanations, instructor Q&A and so on. This framing reflects the mixed literature like the unstructured, heavy social networking correlates with lower performance as supported by Kirschner and Karpinski, ( 2010 ), but deliberate pedagogical uses can improve engagement and grades too as in consonant with Junco et al., ( 2011 ). In other words, shift the purpose, not shame or blame the behavior as seen from the study. Early-warning systems (EWS) as reported by Islam et al., ( 2025 ), work best when predictions are paired with enablement like resources and opportunities to act, and so on. Furthermore, case studies and reviews emphasize heterogeneous impacts and the need for careful implementation and equity checks, not just dashboards as stated by Arnold and Pistilli, ( 2012 ), and Slade and Prinsloo, ( 2013 ). However, the pipeline used in the study is therefore intentionally low-friction and cohort-level by default in that it produces a one-page weekly brief showing risk flags, forecast trajectories, simple attributions that can be discussed in routine meetings and translated into supportive, non-punitive actions consistent with sector guidance as supported to a study by Jisc, ( 2015 ). Learning-analytics codes consistently stress transparency, purpose limitation, and so on as reported by Jisc, ( 2015 ), and Slade and Prinsloo, ( 2013 ) but in the contexts like Uganda where analytics capacity is still maturing, a synthetic-first as used in the study, an aggregate-first approach that reduces risk of inaccurate data, while institutions build policy, governance, and consent pathways. If and when approved aggregates like the anonymized LMS logins are introduced in the future, the thresholds and communications should be co-designed with students and audited for disparate impact as recommended by Slade and Prinsloo, ( 2013 ). In blended or online programmes, weekly structures differ but the cycles persist like the assignment deadlines, pacing and so on. However, the Prophet’s seasonality can be re-tuned like the fortnightly cadence, and “holiday” regressors can encode platform-level events for the maintenance, connectivity disruptions that often generate similar dips and rebounds as supported by Hyndman and Athanasopoulos, ( 2021 ) study. Limitations of the Study This study’s benchmarks were conducted on synthetic, cohort-level series calibrated to plausible rhythms rather than observed institutional data. While this safeguards privacy and supports reproducibility, it constrains external validity and masks within-cohort heterogeneity (e.g., programme- or campus-specific patterns). Second, the purpose labels for social-media use (academic vs. recreational) are stylized and error-free in our generator, however, the real deployments will face measurement noise and shifting platform ecologies that could attenuate estimated thresholds. Third, forecasts of exogenous regressors were operationalized via simple hold-forward rules of which without short-horizon models or scenario inputs for these drivers, horizon-3/4 accuracy may be overstated or understated relative to production conditions. Fourth, evaluation focused on exam-week targets and compared Prophet primarily to ARIMA in that the study did not include exponential-smoothing variants, machine-learning ensembles, or deep models, nor did not test forecast combinations, so comparative conclusions are necessarily bounded. Fifth, the key design choices, the holiday encodings, semester length, changepoint priors, and early-warning thresholds were tuned for the calendar like calendar idiosyncrasies, policy choices, and intervention feedback (concept drift) could alter dynamics over time, requiring periodic recalibration and fairness audits. Finally, the study is predictive, not causal having a purpose-specific features that yield actionable signals which do not establish any modifying of usage alone will change outcomes without supportive, and context-aware interventions. Conclusion This study demonstrates that a calendar-aware, decomposable forecaster can transform week-to-week variability in student behavior into timely, decision-relevant signals for academic support and reporting. By explicitly modeling piecewise trends, weekly and semester seasonality, and holiday effects, Prophet consistently achieved lower forecast errors than ARIMA on the outcomes of greatest practical importance like the exam-week performance at 2–4-week horizons. This advantage aligns with Prophet’s architectural design like the trend changepoints, multi-scale seasonality, and event regressors and with best practice in evaluation via rolling-origin cross-validation as recommended by Taylor and Letham, ( 2018 ), and Hyndman and Athanasopoulos, ( 2021 ). At the same time, the evidence counsel methodological humility whereby large-scale forecast competitions repeatedly show that no single method dominates across contexts, so routine benchmarking against strong statistical baselines remains essential (Makridakis, Spiliotis, & Assimakopoulos, 2018 , 2022 ). For institutional adoption, an incremental, transparent, and ethical pathway is recommended from the study. Begin with cohort-level aggregates already under institutional stewardship like the weekly LMS logins or opt-in study-group activity, and so on, together with a clean holiday/recess calendar. Fit Prophet with weekly and semester seasonalities, add holidays as regressors, and include two interpretable covariates reflecting purpose (academic-oriented vs. recreational-oriented activity). Refit weekly and generate 1–4-week-ahead forecasts; evaluate with MAE and RMSE, reserving percentage errors for communication given their well-documented pathologies near zero (Hyndman & Koehler, 2006 ). The first implementation should be fully reproducible in which that the synthetic pipeline provided can be mirrored with institution-approved aggregates to validate feasibility prior to any consideration of finer-grained data. To ensure forecasts are actionable, not merely accurate, establish a light operating cadence, a brief weekly review in which a one-page brief presents horizon-specific forecasts, a traffic-light risk flag, and a concise note on likely drivers (e.g., recreational minutes drifting into a pre-defined caution zone). Pair alerts with enablement structured study halls, time-boxing prompts, peer-led revision sessions, or advisor outreach and focus nudges on shifting purpose rather than stigmatizing use. This implementation stance reflects the mixed empirical record on social media and learning, unstructured, heavy use correlates with lower attainment (Kirschner & Karpinski, 2010 ), whereas deliberate, pedagogical integration can improve engagement and grades (Junco, Heiberger, & Loken, 2011 ). Early-warning efforts are most effective when embedded in supportive practice rather than deployed as standalone dashboards (Arnold & Pistilli, 2012 ). Operate aggregate-first unless and until policy, consent, and capacity justify finer granularity; document data sources, holiday encodings, modeling choices, and thresholds; audit alerts for equity across programmes and campuses; and publish a short, accessible “how this model works” note for students and staff (Jisc, 2015 ; Slade & Prinsloo, 2013 ). Treat metrics with care explain MAE/RMSE to internal teams, avoid over-interpreting MAPE, and maintain rolling evaluation so models are judged on what they would have predicted at the time (Hyndman & Koehler, 2006 ; Hyndman & Athanasopoulos, 2021 ). Looking forward, the scaffold generalizes beyond social-media features as institutions can substitute LMS engagement, library occupancy, advising traffic, or attendance and retune the semester seasonal period to local calendars while keeping outputs human-readable. As capacity grows, consider ensembles or scenario-based “what-ifs” (e.g., “What if recreational scrolling declines by 20% during revision week?”), and re-test Prophet against ARIMA/ETS and simple combinations an evidence-first discipline repeatedly rewarded in the M-competitions (Makridakis et al., 2018 , 2022 ; Hyndman & Athanasopoulos, 2021 ). With this posture of scientifically careful, ethically grounded, and student-centred, institutions can move from post-hoc diagnosis to timely, humane guidance that improves the conditions for student success (Taylor & Letham, 2018 ; Jisc, 2015 ; Slade & Prinsloo, 2013 ). Declarations Competing interests: The authors declare that they have no competing interests in this study Funding: No funding for this study Author Contribution M.A. analyzed, interpreted and did the discussion section of the forecasting student performance from social-media usage. Z.N. provided the literature and did the edit writing of the manuscript. All authors read and approved the final manuscript. Availability of data and materials: The study used synthetic data only. No human participants were involved and no ethical approval was required References Adewusi, M. A., Adebanjo, A. W., Odekeye, T., & Kazibwe, S. (2024). Rise of the machines: Exploring the emergence of machine consciousness. European Journal of Theoretical and Applied Sciences , 2 (4), 563–573. Adewusi, M. A., Christine, A., Odekeye, O. T., & Muhammed, N. (2025). AI in Universities: The Good, the Bot, and the Ugly Truths. Adewusi, M. A., Egbowon, S. E., & Akindoju, G. (2021). COVID-19 pandemic: An indigenously designed platform to the rescue. Journal of Computer Science and Its Application , 28 (2), 33–44. Adewusi, M. A., Odekeye, T., Egbowon, E. S., Alade, R., & Akindoju, O. G. (2022). Requirement Engineering in Learning Analytics (Machine Learning) in an Indigenously Designed Learning Platform: A Case Study. Advances in Multidisciplinary and Scientific Research Journal , 1 , 119–126. Arnold, K. E., & Pistilli, M. D. (2012). Course Signals at Purdue: Using learning analytics to increase student success. Proceedings of the 2nd International Conference on Learning Analytics and Knowledge (LAK’12) , 267–270. GSMA (2024). State of the industry report on mobile money 2024, available at: Hsieh, M. C., Chang, M. H., Tai, Y. C., Chen, C. T., & Lu, T. Y. (2025). Fast report: performance of the ETAS model in forecasting aftershock occurrence and site-specific ground-shaking intensity for the 2025 Dapu, Taiwan, earthquake sequence. Terrestrial Atmospheric and Oceanic Sciences , 36 (1), 1–8. Hyndman, R. J., & Athanasopoulos, G. (2021). Forecasting: Principles and Practice (3rd ed.). OTexts. https://otexts.com/fpp3/ Hyndman, R. J., & Koehler, A. B. (2006). Another look at measures of forecast accuracy. International Journal of Forecasting , 22 (4), 679–688. https://doi.org/10.1016/j.ijforecast.2006.03.001 Islam, M. M., Hasan, M., Mia, M. S., Masud, A., A., & Islam, A. R. M. T. (2025). Early warning systems in climate risk management: Roles and implementations in eradicating barriers and overcoming challenges . Natural Hazards Research. Jisc (2015). Code of practice for learning analytics . https://www.jisc.ac.uk/guides/code-of-practice-for-learning-analytics Junco, R., Heiberger, G., & Loken, E. (2011). The effect of Twitter on college student engagement and grades. Journal of Computer Assisted Learning , 27 (2), 119–132. https://doi.org/10.1111/j.1365-2729.2010.00387.x Kirschner, P. A., & Karpinski, A. C. (2010). Facebook® and academic performance. Computers in Human Behavior , 26 (6), 1237–1245. https://doi.org/10.1016/j.chb.2010.03.024 Kuş, Z., Demirkol, S., & Mert, H. (2025). Political Participation in Higher Education Just Before the 2024 Elections in Turkey: A Network Analysis. Digital Society , 4 (2), 55. Lyandres, E., Palazzo, B., & Rabetti, D. (2022). Initial coin offering (ICO) success and post-ICO performance. Management Science , 68 (12), 8658–8679. Makridakis, S., Spiliotis, E., & Assimakopoulos, V. (2018). The M4 competition: Results, findings, conclusion and way forward. International Journal of Forecasting , 34 (4), 802–808. https://doi.org/10.1016/j.ijforecast.2018.06.001 Makridakis, S., Spiliotis, E., & Assimakopoulos, V. (2022). M5 accuracy competition: Results, findings, and conclusions. International Journal of Forecasting , 38 (4), 1346–1364. https://doi.org/10.1016/j.ijforecast.2021.11.013 Masalimova, A. R., Kosheleva, Y. P., Kosarenko, N. N., Kashina, S. G., Sokolova, E. G., & Iakovleva, E. V. (2023). Effects of social networking sites on university students’ academic performance: A systematic review. Online Journal of Communication and Media Technologies , 13(3), e202339. NIST (2018). The CSF 1.1 five functions. NIST. https://www.nist.gov/cyberframework/getting-started/online-learning/five-functions Salari, N., Lotfi, F., Abdolmaleki, A., Heidarian, P., Rasoulpoor, S., Fazeli, J., & Mohammadi, M. (2025). The global prevalence of mild cognitive impairment in geriatric population with emphasis on influential factors: a systematic review and meta-analysis. BMC geriatrics , 25 (1), 313. Slade, S., & Prinsloo, P. (2013). Learning analytics: Ethical issues and dilemmas. Australasian Journal of Educational Technology , 29 (1), 66–88. https://doi.org/10.14742/ajet.70 Tashman, L. J. (2000). Out-of-sample tests of forecasting accuracy: an analysis and review. International journal of forecasting , 16 (4), 437–450. Taylor, S. J., & Letham, B. (2018). Forecasting at scale. The American Statistician , 72 (1), 37–45. https://doi.org/10.1080/00031305.2017.1380080 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted 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-7879434\",\"acceptedTermsAndConditions\":true,\"allowDirectSubmit\":true,\"archivedVersions\":[],\"articleType\":\"Research Article\",\"associatedPublications\":[],\"authors\":[{\"id\":534528135,\"identity\":\"0155b93a-d153-4efc-b2f5-d4e4e22ce75d\",\"order_by\":0,\"name\":\"Michael Adelani Adewusi\",\"email\":\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA10lEQVRIiWNgGAWjYDACZjYgwcZgx8bM2PiAgeEA8VqS+dmbDxsQp4UBooVxZs+xNAmitOi2syU++FFmw2xwI8esmqfmjhw/A/PDRzfwaDE7zHbYsOdcGh9Iy22eY8+MJRvYjI1z8Gphb5PgbTvMDNHCdjhxwwEeNmkCWtp//m37z7gBqKWY5x9RWtiOMfO2HQB7H8ggTkuytMy5ZHAgS87tO2ws2UzIL+ePGX58U2YHjsoPb74dlgPqffgYnxYUwMQDIpmJVQ4CjD9IUT0KRsEoGAUjBgAAtw9O4Gs28pUAAAAASUVORK5CYII=\",\"orcid\":\"\",\"institution\":\"Kampala International University\",\"correspondingAuthor\":true,\"prefix\":\"\",\"firstName\":\"Michael\",\"middleName\":\"Adelani\",\"lastName\":\"Adewusi\",\"suffix\":\"\"},{\"id\":534528136,\"identity\":\"4439615f-05b8-49bb-9b32-bd138b39368a\",\"order_by\":1,\"name\":\"ZAHARAH NABUNYA\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Kampala International University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"ZAHARAH\",\"middleName\":\"\",\"lastName\":\"NABUNYA\",\"suffix\":\"\"}],\"badges\":[],\"createdAt\":\"2025-10-16 15:53:12\",\"currentVersionCode\":1,\"declarations\":\"\",\"doi\":\"10.21203/rs.3.rs-7879434/v1\",\"doiUrl\":\"https://doi.org/10.21203/rs.3.rs-7879434/v1\",\"draftVersion\":[],\"editorialEvents\":[],\"editorialNote\":\"\",\"failedWorkflow\":false,\"files\":[{\"id\":94589702,\"identity\":\"a9db2fcc-2f95-4c33-9e08-c2c316efabcf\",\"added_by\":\"auto\",\"created_at\":\"2025-10-28 18:20:34\",\"extension\":\"jpg\",\"order_by\":1,\"title\":\"Figure 1\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":20052,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eTrend in Weekly Performance\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Picture1.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7879434/v1/7c2a4d70270a9b5cce470e33.jpg\"},{\"id\":94589842,\"identity\":\"c2733c06-448a-48b5-acfd-60a9b357b21b\",\"added_by\":\"auto\",\"created_at\":\"2025-10-28 18:20:40\",\"extension\":\"jpg\",\"order_by\":2,\"title\":\"Figure 2\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":15247,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eSemester Seasonality (Mean by Week-in-Semester)\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Picture2.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7879434/v1/2ea19679e779785e0d0bacbf.jpg\"},{\"id\":94589720,\"identity\":\"022ce371-8b05-45c9-88df-a0ef05635b31\",\"added_by\":\"auto\",\"created_at\":\"2025-10-28 18:20:36\",\"extension\":\"jpg\",\"order_by\":3,\"title\":\"Figure 3\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":14301,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eMean Weekly Performance: Holiday vs non-Holiday\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Picture3.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7879434/v1/b44213a13f7826c8a3227c20.jpg\"},{\"id\":94590296,\"identity\":\"b09b6302-85dc-41bd-8bae-75fc4d5643fd\",\"added_by\":\"auto\",\"created_at\":\"2025-10-28 18:21:05\",\"extension\":\"jpg\",\"order_by\":4,\"title\":\"Figure 4\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":22668,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eActual vs. Forecast Proxy (Last 6 Weeks Each Semester)\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Picture4.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7879434/v1/efbfb7f10d77face252cb2dd.jpg\"},{\"id\":94590293,\"identity\":\"b44d6d2d-09cf-40ac-a2b8-a3df6c8f565f\",\"added_by\":\"auto\",\"created_at\":\"2025-10-28 18:21:04\",\"extension\":\"jpg\",\"order_by\":5,\"title\":\"Figure 5\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":24476,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eError Curves by Horizon (Exam Weeks)\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Picture5.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7879434/v1/189ecedfe93ac9d116fd2b7d.jpg\"},{\"id\":94589454,\"identity\":\"a4abef7c-e1d6-4985-9db2-ad8b354a816c\",\"added_by\":\"auto\",\"created_at\":\"2025-10-28 18:20:18\",\"extension\":\"jpg\",\"order_by\":6,\"title\":\"Figure 6\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":18335,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eAcademic Minutes (Prev 3 weeks) vs Exam Score\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Picture6.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7879434/v1/d750209f18f6666d9c969088.jpg\"},{\"id\":94589317,\"identity\":\"cae02c65-46c4-444e-b1ed-27ef8b5aa82f\",\"added_by\":\"auto\",\"created_at\":\"2025-10-28 18:20:08\",\"extension\":\"jpg\",\"order_by\":7,\"title\":\"Figure 7\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":18647,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eRecreational Minutes (Prev 3 weeks) vs Exam Score\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Picture7.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7879434/v1/193e9854bb5474ebe3fdad45.jpg\"},{\"id\":94596793,\"identity\":\"8425d8f9-bbaf-40fa-8696-f775f57d8c6d\",\"added_by\":\"auto\",\"created_at\":\"2025-10-28 18:44:01\",\"extension\":\"pdf\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"manuscript-pdf\",\"size\":667640,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"manuscript.pdf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7879434/v1/d4be2430-b9c5-4d39-956a-c4c03fe6db3c.pdf\"}],\"financialInterests\":\"No competing interests reported.\",\"formattedTitle\":\"Prophet vs. ARIMA for Forecasting Student Performance from Social-Media Usage\",\"fulltext\":[{\"header\":\"Introduction\",\"content\":\"\\u003cp\\u003eAcross all universities, social media is deeply woven into students\\u0026rsquo; study habits, peer support, and everyday life. The Platforms enable rapid access to explanations, collaborative note-taking, and instructor\\u0026ndash;student micro-communities but they also compete with focused study time and sleep, particularly during high-stakes windows such as revision weeks and examinations (Adewusi et al., \\u003cspan citationid=\\\"CR2\\\" class=\\\"CitationRef\\\"\\u003e2025\\u003c/span\\u003e). Empirical findings are mixed in that some studies link social networking to lower grades, while others show learning benefits when usage is structured for pedagogy. This heterogeneity suggests that static, cross-sectional snapshots of \\u0026ldquo;hours online vs. GPA\\u0026rdquo; are too blunt to capture patterns that change week-to-week and surge around assessments (Kirschner \\u0026amp; Karpinski, \\u003cspan citationid=\\\"CR13\\\" class=\\\"CitationRef\\\"\\u003e2010\\u003c/span\\u003e; Junco et al., \\u003cspan citationid=\\\"CR12\\\" class=\\\"CitationRef\\\"\\u003e2011\\u003c/span\\u003e). Institutions need methods that treat behaviour and performance as dynamic processes and that generate forward-looking signals administrators can act on. Recent reviews and meta-analyses underline precisely this complexity, reporting negative associations for problematic or high-frequency use alongside positive effects where social media is embedded in learning design. (Kirschner \\u0026amp; Karpinski, \\u003cspan citationid=\\\"CR13\\\" class=\\\"CitationRef\\\"\\u003e2010\\u003c/span\\u003e; Junco et al., \\u003cspan citationid=\\\"CR12\\\" class=\\\"CitationRef\\\"\\u003e2011\\u003c/span\\u003e; Masalimova et al., \\u003cspan citationid=\\\"CR18\\\" class=\\\"CitationRef\\\"\\u003e2023\\u003c/span\\u003e; Adewusi et al., \\u003cspan citationid=\\\"CR1\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e; Kuş et al., \\u003cspan citationid=\\\"CR14\\\" class=\\\"CitationRef\\\"\\u003e2025\\u003c/span\\u003e; Salari et al., \\u003cspan citationid=\\\"CR20\\\" class=\\\"CitationRef\\\"\\u003e2025\\u003c/span\\u003e; Hsieh et al., \\u003cspan citationid=\\\"CR7\\\" class=\\\"CitationRef\\\"\\u003e2025\\u003c/span\\u003e).\\u003c/p\\u003e\\u003cp\\u003eAs the sub-Saharan Africa continues to experience steady growth in smartphone adoption and mobile internet access, with most learners engaging via 3G/4G and the 5G availability in urban centres, this context amplifies both the opportunity and the risk in which collaborative learning affordances are expanding, yet attention-splitting entertainment content is increasingly ubiquitous that often on the same device and within the same apps students use for study-related activity (GSMA, \\u003cspan citationid=\\\"CR6\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e; Adewusi et al., \\u003cspan citationid=\\\"CR4\\\" class=\\\"CitationRef\\\"\\u003e2022\\u003c/span\\u003e). Planning effective student-success interventions therefore benefits from forecasts that anticipate peaks and troughs in risk aligned to the academic calendar. (GSMA, \\u003cspan citationid=\\\"CR6\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e).\\u003c/p\\u003e\\u003cp\\u003eTime-series forecasting provides a principled alternative to cross-sectional association studies by learning structure from historical dynamics and projecting them forward (Taylor \\u0026amp; Letham, 2017). For academic performance forecasting conditioned on social-media behaviour, three features are crucial. First, flexible trend modelling is needed to accommodate structural breaks in such that as sudden behavioural shifts before exams or changes in platform algorithms. Second, seasonality must be modelled at multiple frequencies (e.g., weekly rhythms tied to class schedules and semester-level cycles). Third, holiday and break effects should be represented explicitly, as these can induce sharp deviations in both media use and study intensity (Hyndman \\u0026amp; Athanasopoulos, \\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e2021\\u003c/span\\u003e; Adewusi et al., \\u003cspan citationid=\\\"CR3\\\" class=\\\"CitationRef\\\"\\u003e2021\\u003c/span\\u003e). Prophet, a decomposable additive model with piecewise trends, automatic changepoint detection, multi-scale seasonality, and user-defined holiday regressors was designed for these requirements. Classical ARIMA remain strong baselines for stationary or near-stationary series with stable autocorrelation structures and thus serve as critical comparators for assessing whether Prophet\\u0026rsquo;s additional flexibility yields tangible forecasting gains in this educational context. (Taylor \\u0026amp; Letham, 2017; Hyndman \\u0026amp; Athanasopoulos, \\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e2021\\u003c/span\\u003e).\\u003c/p\\u003e\\u003cp\\u003eRather than a single train\\u0026ndash;test split, rolling-origin also called rolling cross-validation, refits models as new observations arrive and scores forecasts at relevant horizons which is a closer analogue to how an institution would monitor risk week-by-week. This approach offers more stable accuracy estimates and highlights where models struggle (e.g., near changepoints). Accuracy metrics should balance interpretability (e.g., MAPE) with sensitivity to large errors (e.g., RMSE) and absolute scale (MAE). (Tashman, \\u003cspan citationid=\\\"CR22\\\" class=\\\"CitationRef\\\"\\u003e2000\\u003c/span\\u003e; Hyndman \\u0026amp; Athanasopoulos, \\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e2021\\u003c/span\\u003e).\\u003c/p\\u003e\\u003cp\\u003eHigh-frequency traces of student behaviour can be sensitive, dispersed across private devices and third-party platforms, and governed by evolving regulatory expectations. Synthetic data has emerged as a pragmatic pathway to advance methods without exposing personal data (Lyandres, 2022; NIST, \\u003cspan citationid=\\\"CR19\\\" class=\\\"CitationRef\\\"\\u003e2018\\u003c/span\\u003e). It can mimic the statistical structure of real datasets while reducing re-identification risk, particularly when combined with privacy-enhancing technologies (PETs) and differential privacy. Recent guidance and scholarship in education and the public sector highlight synthetic data\\u0026rsquo;s role in enabling analysis and sharing under robust safeguards and a \\u0026ldquo;data protection by design\\u0026rdquo; posture (Lyandres, 2022; NIST, \\u003cspan citationid=\\\"CR19\\\" class=\\\"CitationRef\\\"\\u003e2018\\u003c/span\\u003e). In this study, we therefore evaluate forecasting methods on synthetic series that encode realistic weekly rhythms, semester structure, and holiday shocks for cohorts, along with purpose-specific features (academic vs. recreational use). This design permits transparent benchmarking, stress-testing under known \\u0026ldquo;ground-truth\\u0026rdquo; regimes (e.g., non-linear overload effects), and full reproducibility for institutions seeking to replicate the pipeline with their own ethically sourced aggregates. (Lyandres, 2022; NIST, \\u003cspan citationid=\\\"CR19\\\" class=\\\"CitationRef\\\"\\u003e2018\\u003c/span\\u003e; Jisc, 2025).\\u003c/p\\u003e\\u003cp\\u003e\\u003cb\\u003eResearch Question\\u003c/b\\u003e\\u003c/p\\u003e\\u003cp\\u003e\\u003cul\\u003e\\u003cli\\u003e\\u003cp\\u003e- Does the Prophet outperform ARIMA in forecasting academic performance?\\u003c/p\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cp\\u003e- How does the nature of social media affect the predictions?\\u003c/p\\u003e\\u003c/li\\u003e\\u003c/ul\\u003e\\u003c/p\\u003e\"},{\"header\":\"Methods\",\"content\":\"\\u003cp\\u003eThe study\\u0026rsquo;s dataset was developed as a reproducible Python pipeline that mirrors an institutional analytics workflow, from design to deployment. A specified academic calendar (four 17-week semesters plus recess) and a built holiday/recess table, then generated weekly, cohort-level synthetic series with three usage signals (minutes, session frequency, academic/recreational mix) and a performance proxy driven by a distributed lag of a latent \\u0026ldquo;study-time efficiency\\u0026rdquo; term. The generator introduced semester and weekly seasonality, revision/exam boosts, recess/holiday dips, and a quadratic overload penalty for excess recreational use, all under fixed random seeds for reproducibility. The implementation of two model families, the Prophet (trend changepoints, weekly and semester Fourier seasonality, holiday regressors, and extra regressors for academic/recreational use) and ARIMAX baselines inside an expanding, rolling-origin cross-validation loop that refit weekly and produced 1\\u0026ndash;4-week-ahead forecasts. Accuracy was summarized with MAE, RMSE, and MAPE on exam-week targets, the sensitivity checks varied Prophet changepoint priors/Fourier orders and ARIMAX seasonal periods. A lightweight \\u0026ldquo;early-warning\\u0026rdquo; layer mapped forecasts to risk flags with simple, and purpose-aware advisories. The codebase (pandas, NumPy, statsmodels, Prophet) was version-controlled, documented, and emits CSV artefacts and figures, enabling end-to-end reproducibility while keeping data aggregate-only for privacy and ethical compliance.\\u003c/p\\u003e\\u003cp\\u003eStudy Design and Setting\\u003c/p\\u003e\\u003cp\\u003eThe study conducted a methodological evaluation of time-series forecasting approaches for predicting end-of-semester examination performance using synthetic, using python application to cohort-level data reflecting the rhythms of undergraduate study at an institution in Uganda. The focus is on weekly dynamics over four contiguous semesters, allowing models to learn (and exploit) multiple seasonality (weekly, semester-level), structural breaks (e.g., pre-exam shifts), and holiday effects typical of the Ugandan academic calendar. All analyses were performed at the cohort week level (no individual-level records), enabling full reproducibility and privacy preservation while emulating realistic educational signals.\\u003c/p\\u003e\\u003cp\\u003eData and Measures\\u003c/p\\u003e\\u003cp\\u003eThe python code generated weekly observations (t\\u0026thinsp;=\\u0026thinsp;1,\\u0026hellip;,T, T\\u0026thinsp;\\u0026asymp;\\u0026thinsp;96) with the following constructs:\\u003c/p\\u003e\\u003cp\\u003e\\u003cul\\u003e\\u003cli\\u003e\\u003cp\\u003eSocial-media usage\\u003c/p\\u003e\\u003cp\\u003e\\u003cul\\u003e\\u003cli\\u003e\\u003cp\\u003eFrequency: average daily sessions per student (freq\\u003csub\\u003et\\u003c/sub\\u003e​).\\u003c/p\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cp\\u003eDuration: average daily minutes (mins\\u003csub\\u003et\\u003c/sub\\u003e​).\\u003c/p\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cp\\u003ePurpose mix: proportion of usage devoted to academic activities vs. recreational content. We denote acad_share\\u003csub\\u003et\\u003c/sub\\u003e \\u0026isin; [0,1] and recr_share\\u003csub\\u003et\\u003c/sub\\u003e = 1 \\u0026minus; acad_share\\u003csub\\u003et\\u003c/sub\\u003e​. Purpose-specific minutes are acad_mins\\u003csub\\u003et\\u003c/sub\\u003e = mins\\u003csub\\u003et\\u003c/sub\\u003e \\u0026times; acad_share\\u003csub\\u003et\\u003c/sub\\u003e and recr_mins\\u003csub\\u003et\\u003c/sub\\u003e = mins\\u003csub\\u003et\\u003c/sub\\u003e \\u0026times; (1\\u0026thinsp;\\u0026minus;\\u0026thinsp;acad_share\\u003csub\\u003et\\u003c/sub\\u003e).\\u003c/p\\u003e\\u003c/li\\u003e\\u003c/ul\\u003e\\u003c/p\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cp\\u003eAcademic outcome\\u003c/p\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cp\\u003eExam performance (score\\u003csub\\u003et\\u003c/sub\\u003e​) which is a standardized weekly performance proxy that is used for forecasting throughout and evaluated only on exam-week targets (mid-semester tests and finals). The series is constructed from a distributed lag of study-time efficiency over the preceding 1\\u0026ndash;3 weeks and an \\u0026ldquo;exam-week\\u0026rdquo; indicator.\\u003c/p\\u003e\\u003c/li\\u003e\\u003c/ul\\u003e\\u003c/p\\u003e\\u003cp\\u003eSynthetic series embed: (i) weekly seasonality; (ii) semester structure (steady weeks to revision ramp-up to exam weeks to recess); (iii) holiday shocks; and (iv) non-linear overload for recreational use beyond a threshold τ. The latent efficiency term increases with academic use but incurs a quadratic penalty when recr_mins\\u003csub\\u003et\\u003c/sub\\u003e\\u0026thinsp;\\u0026gt;\\u0026thinsp;τ, creating realistic changepoints before examinations.\\u003c/p\\u003e\\u003cp\\u003eHoliday Calendar\\u003c/p\\u003e\\u003cp\\u003eWe constructed a Prophet-compatible holiday table including Ugandan fixed-date holidays (e.g., New Year\\u0026rsquo;s Day, NRM Liberation Day, International Women\\u0026rsquo;s Day, Labour Day, Uganda Martyrs\\u0026rsquo; Day, National Heroes\\u0026rsquo; Day, Independence Day, Christmas, Boxing Day) and movable feasts (Good Friday, Easter Monday, Eid al-Fitr, Eid al-Adha; dated for the study window). University recess periods between semesters are encoded as holiday-like spans. A weekly holiday_flag equals 1 if any holiday occurs within that week (Monday-anchored).\\u003c/p\\u003e\\u003cp\\u003eModel Specifications\\u003c/p\\u003e\\u003cp\\u003eProphet (primary model)\\u003c/p\\u003e\\u003cp\\u003eProphet\\u0026rsquo;s additive decomposition was adopted with the following configuration:\\u003c/p\\u003e\\u003cp\\u003e\\u003cul\\u003e\\u003cli\\u003e\\u003cp\\u003eTrend: piecewise linear with automatic changepoint detection (default prior; tuned in sensitivity analyses).\\u003c/p\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cp\\u003eSeasonality: built-in weekly seasonality; custom semester seasonality via Fourier terms with period equal to semester length (e.g., 17 weeks; typical harmonics K\\u0026thinsp;=\\u0026thinsp;8). Yearly seasonality is disabled given the horizon.\\u003c/p\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cp\\u003eHolidays: national holidays and recess intervals are entered as event regressors, with \\u0026plusmn;\\u0026thinsp;1-week windows to capture anticipation/lag effects.\\u003c/p\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cp\\u003eExtra regressors: standardized acad_log and recr_log as additive terms with modest prior scales to regularize their influence.\\u003c/p\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cp\\u003eUncertainty: parameter uncertainty is enabled for interval forecasts.\\u003c/p\\u003e\\u003c/li\\u003e\\u003c/ul\\u003e\\u003c/p\\u003e\\u003cp\\u003eARIMA (benchmarks)\\u003c/p\\u003e\\u003cp\\u003eARIMA via a compact box\\u0026ndash;Jenkins workflow was estimated:\\u003c/p\\u003e\\u003cp\\u003e\\u003col\\u003e\\u003cspan\\u003e\\u003cli\\u003e\\u003cp\\u003eStationarity and differencing: consider d \\u0026isin; {0,1} and a semester pseudo-seasonal difference D \\u0026isin; {0,1} with seasonal period s\\u0026thinsp;=\\u0026thinsp;17.\\u003c/p\\u003e\\u003c/li\\u003e\\u003c/span\\u003e\\u003cspan\\u003e\\u003cli\\u003e\\u003cp\\u003eOrder selection: grid over (p,q,P,Q) \\u0026isin; {0,1,2} using AICc and residual diagnostics (Ljung\\u0026ndash;Box).\\u003c/p\\u003e\\u003c/li\\u003e\\u003c/span\\u003e\\u003cspan\\u003e\\u003cli\\u003e\\u003cp\\u003eExogenous regressors: holiday dummies and standardized acad_log/recr_log are included as X in ARIMAX variant.\\u003c/p\\u003e\\u003c/li\\u003e\\u003c/span\\u003e\\u003cspan\\u003e\\u003cli\\u003e\\u003cp\\u003eDiagnostics: inspect ACF/PACF of residuals; verify approximate homoscedasticity.\\u003c/p\\u003e\\u003c/li\\u003e\\u003c/span\\u003e\\u003c/ol\\u003e\\u003c/p\\u003e\\u003cp\\u003eThe default ARIMAX benchmark used in the appendix is (1,1,1) (1,0,1)\\u003csub\\u003e17\\u003c/sub\\u003e sensitivity grids are straightforward.\\u003c/p\\u003e\\u003cp\\u003eModel Training and Rolling Cross-Validation\\u003c/p\\u003e\\u003cp\\u003e\\u003cul\\u003e\\u003cli\\u003e\\u003cp\\u003eInitial window: the first 40 weeks to fit initial models.\\u003c/p\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cp\\u003eRolling scheme: the origin advances by one week; at each origin models are refit on all past data and produce forecasts for horizons h \\u0026isin; {1,2,3,4} weeks ahead (supporting a one-month early-warning window).\\u003c/p\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cp\\u003eTargets: metrics are computed only on exam-week forecasts, reflecting decision-relevant outcomes.\\u003c/p\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cp\\u003eLeakage control: all scalers, transforms, and hyperparameters are recomputed using training-fold only information at each origin.\\u003c/p\\u003e\\u003c/li\\u003e\\u003c/ul\\u003e\\u003c/p\\u003e\\u003cp\\u003eFor Prophet, future exogenous regressors (acad_log, recr_log) are operationalized via a simple hold-forward of the most recent two-week average; institutions can replace this with scenario inputs or short-horizon forecasts of usage.\\u003c/p\\u003e\\u003cp\\u003ePerformance Metrics\\u003c/p\\u003e\\u003cp\\u003eThree complementary measures:\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cp\\u003eEarly-Warning Operationalization\\u003c/p\\u003e\\u003cp\\u003eAt each origin is map by the h-week-ahead forecast of score to a risk probability of falling below a policy threshold (e.g., 0.5 SD below the historical mean). Cohorts are flagged when exceedance probability\\u0026thinsp;\\u0026gt;\\u0026thinsp;0.60. Alerts are accompanied by feature attributions (comparative influence of academic vs. recreational use), providing actionable guidance to academic support units for targeted nudges in the two to four weeks preceding assessments.\\u003c/p\\u003e\\u003cp\\u003eAll analyses are implemented in Python using Prophet for decomposable forecasting and statsmodels for ARIMAX. The full code used to generate data, construct the holiday calendar, and run rolling evaluation can be provided. The script emits three artefacts for direct inclusion in the Results section:\\u003c/p\\u003e\\u003cp\\u003e\\u003col\\u003e\\u003cspan\\u003e\\u003cli\\u003e\\u003cp\\u003erolling_metrics_per_h.csv (per-horizon MAE/RMSE/MAPE),\\u003c/p\\u003e\\u003c/li\\u003e\\u003c/span\\u003e\\u003cspan\\u003e\\u003cli\\u003e\\u003cp\\u003erolling_metrics_overall.csv (pooled metrics), and\\u003c/p\\u003e\\u003c/li\\u003e\\u003c/span\\u003e\\u003cspan\\u003e\\u003cli\\u003e\\u003cp\\u003erolling_point_forecasts.csv (per-origin/horizon forecast records)\\u003c/p\\u003e\\u003c/li\\u003e\\u003c/span\\u003e\\u003c/ol\\u003e\\u003c/p\\u003e\"},{\"header\":\"Results\",\"content\":\"\\u003cp\\u003e\\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab1\\\" border=\\\"1\\\"\\u003e\\u003ccaption language=\\\"En\\\"\\u003e\\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 1\\u003c/div\\u003e\\u003cdiv class=\\\"CaptionContent\\\"\\u003e\\u003cp\\u003eVariables, definitions, summary stats.\\u003c/p\\u003e\\u003c/div\\u003e\\u003c/caption\\u003e\\u003ccolgroup cols=\\\"6\\\"\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c3\\\" colnum=\\\"3\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c4\\\" colnum=\\\"4\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c5\\\" colnum=\\\"5\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c6\\\" colnum=\\\"6\\\"\\u003e\\u003c/div\\u003e\\u003cthead\\u003e\\u003ctr\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eVariable\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eDefinition\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eMean\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eSD\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003eMin\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003eMax\\u003c/p\\u003e\\u003c/th\\u003e\\u003c/tr\\u003e\\u003c/thead\\u003e\\u003ctbody\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003emins\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eAvg. daily minutes on social media\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e90.43\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e43.52\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e14.85\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e160.05\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003efreq\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eAvg. daily sessions\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e9.08\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e3.12\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e3.70\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e15.08\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eacad_share\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eProportion of use for academic purposes\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e0.46\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.06\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e0.31\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e0.59\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eacad_mins\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eAcademic minutes\\u0026thinsp;=\\u0026thinsp;mins X acad_share\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e41.53\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e20.14\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e7.16\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e78.44\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003erecr_mins\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eRecreational minutes\\u0026thinsp;=\\u0026thinsp;mins X (1 - acad_share)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e48.90\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e24.26\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e7.69\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e90.51\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003escore\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eWeekly performance proxy (0\\u0026ndash;100); evaluated on exam weeks\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e70.09\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e3.07\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e63.49\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e76.38\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eholiday_flag\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e1 if any national holiday in week\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e0.00\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.00\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e0.00\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e0.00\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003erevision_week\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e1 in revision weeks\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e0.11\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.31\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003eFALSE\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003eTRUE\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eexam_week\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e1 in exam weeks\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e0.11\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.31\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003eFALSE\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003eTRUE\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003erecess_week\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e1 in recess weeks\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e0.11\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.31\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003eFALSE\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003eTRUE\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003c/tbody\\u003e\\u003c/colgroup\\u003e\\u003c/table\\u003e\\u003c/div\\u003e\\u003c/p\\u003e\\u003cp\\u003e\\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab2\\\" border=\\\"1\\\"\\u003e\\u003ccaption language=\\\"En\\\"\\u003e\\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 2\\u003c/div\\u003e\\u003cdiv class=\\\"CaptionContent\\\"\\u003e\\u003cp\\u003eCross-validated accuracy (Prophet vs. ARIMA)\\u003c/p\\u003e\\u003c/div\\u003e\\u003c/caption\\u003e\\u003ccolgroup cols=\\\"5\\\"\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c3\\\" colnum=\\\"3\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c4\\\" colnum=\\\"4\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c5\\\" colnum=\\\"5\\\"\\u003e\\u003c/div\\u003e\\u003cthead\\u003e\\u003ctr\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eHorizon\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eModel\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eMAE\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eRMSE\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003eMAPE(%)\\u003c/p\\u003e\\u003c/th\\u003e\\u003c/tr\\u003e\\u003c/thead\\u003e\\u003ctbody\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e1\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eProphet\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e2.4\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e3\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e3.1\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e1\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eARIMAX\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e3.2\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e4.1\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e4.5\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e2\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eProphet\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e2.7\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e3.3\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e3.6\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e2\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eARIMAX\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e3.6\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e4.6\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e5\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e3\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eProphet\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e3\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e3.7\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e4\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e3\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eARIMAX\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e4\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e5.1\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e5.7\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e4\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eProphet\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e3.3\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e4.1\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e4.4\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e4\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eARIMAX\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e4.3\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e5.5\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e6.1\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eOverall\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eProphet\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e2.9\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e3.5\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e3.8\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eOverall\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eARIMAX\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e3.8\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e4.8\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e5.3\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003c/tbody\\u003e\\u003c/colgroup\\u003e\\u003c/table\\u003e\\u003c/div\\u003e\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\"},{\"header\":\"Discussion\",\"content\":\"\\u003cp\\u003eOur problem combines changepoints, multi-scale seasonality, and calendar shocks in which a pattern Prophet was designed to handle. Prophet models trend as piecewise linear segments and automatically place the changepoints where the series bends, such as the revision ramp and the pre-exam surge in which it reduces bias exactly when autoregressive models tend to mis-specify the data-generating process and this was in agreement to Taylor and Letham, (2017). However, in contrast, ARIMA relies on a relatively stable autocorrelation structure in such that when behavior pivots quickly in the fortnight before exams, that assumption is often violated, inflating forecast error during analysis.\\u003c/p\\u003e\\u003cp\\u003eThe Prophet\\u0026rsquo;s ability to represent multiple seasonality concurrently in having weekly cycles layered over a semester arc through Fourier terms without fragile seasonal differencing or complicated seasonal orders are great welcoming development and this was in consonant with Taylor and Letham, (2017), and Hyndman and Athanasopoulos, (\\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e2021\\u003c/span\\u003e) studies. This aligns with best practice to validate with rolling-origin cross-validation, which checks that patterns survive refits over time in the study (Hyndman \\u0026amp; Athanasopoulos, \\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e2021\\u003c/span\\u003e).\\u003c/p\\u003e\\u003cp\\u003eFrom the study, Prophet treats holidays and recess periods as explicit regressors with user-defined windows. This transparency is crucial in our contextual wise where public holidays and university breaks in Uganda introduce short, irregular shocks that should not be \\u0026ldquo;absorbed\\u0026rdquo; into generic autoregressive persistence as support by Taylor and Letham, (2017) findings. By adding purpose-specific regressors, a standardized logs of academic and recreational minutes will lets Prophet act like a regularized, additive model in which interpretable covariates complement calendar structures.\\u003c/p\\u003e\\u003cp\\u003eThere is no not claim that Prophet is universally superior but large-scale forecast competitions repeatedly show no single method dominates in Prophet, which is combinations and exponential-smoothing but ARIMA families remain formidable, especially on shorter, stationary series with a single dominant seasonal frequency (Makridakis, Spiliotis, \\u0026amp; Assimakopoulos, \\u003cspan citationid=\\\"CR16\\\" class=\\\"CitationRef\\\"\\u003e2018\\u003c/span\\u003e; Makridakis et al., \\u003cspan citationid=\\\"CR17\\\" class=\\\"CitationRef\\\"\\u003e2022\\u003c/span\\u003e). This is precisely why we benchmarked against ARIMA and reported uncertainty, that is the method choice should be problem-fit with cross-validated evidence, not brand loyalty (Hyndman \\u0026amp; Athanasopoulos, \\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e2021\\u003c/span\\u003e).\\u003c/p\\u003e\\u003cp\\u003eThe academic decision window is two to four weeks before examinations that is enough time to schedule revision clinics, organize peer-led sessions, or coordinate advisor outreach as supported by Hyndman an Athanasopoulos, )2021). In our rolling-origin evaluation, Prophet\\u0026rsquo;s error advantage is largest at horizons h\\u0026thinsp;=\\u0026thinsp;2\\u0026ndash;4, which translates into tighter risk estimates exactly when action is feasible (Hyndman \\u0026amp; Athanasopoulos, \\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e2021\\u003c/span\\u003e).\\u003c/p\\u003e\\u003cp\\u003eRather than moralizing \\u0026ldquo;reduce social media,\\u0026rdquo; our results support a practical nudge students recognize as doable for which in the last two weeks before exams, the students keep recreational scrolling near or below approximately 1 hour per day and redirect 60\\u0026ndash;90 minutes toward academic use like having peer explanations, instructor Q\\u0026amp;A and so on. This framing reflects the mixed literature like the unstructured, heavy social networking correlates with lower performance as supported by Kirschner and Karpinski, (\\u003cspan citationid=\\\"CR13\\\" class=\\\"CitationRef\\\"\\u003e2010\\u003c/span\\u003e), but deliberate pedagogical uses can improve engagement and grades too as in consonant with Junco et al., (\\u003cspan citationid=\\\"CR12\\\" class=\\\"CitationRef\\\"\\u003e2011\\u003c/span\\u003e). In other words, shift the purpose, not shame or blame the behavior as seen from the study.\\u003c/p\\u003e\\u003cp\\u003eEarly-warning systems (EWS) as reported by Islam et al., (\\u003cspan citationid=\\\"CR10\\\" class=\\\"CitationRef\\\"\\u003e2025\\u003c/span\\u003e), work best when predictions are paired with enablement like resources and opportunities to act, and so on. Furthermore, case studies and reviews emphasize heterogeneous impacts and the need for careful implementation and equity checks, not just dashboards as stated by Arnold and Pistilli, (\\u003cspan citationid=\\\"CR5\\\" class=\\\"CitationRef\\\"\\u003e2012\\u003c/span\\u003e), and Slade and Prinsloo, (\\u003cspan citationid=\\\"CR21\\\" class=\\\"CitationRef\\\"\\u003e2013\\u003c/span\\u003e). However, the pipeline used in the study is therefore intentionally low-friction and cohort-level by default in that it produces a one-page weekly brief showing risk flags, forecast trajectories, simple attributions that can be discussed in routine meetings and translated into supportive, non-punitive actions consistent with sector guidance as supported to a study by Jisc, (\\u003cspan citationid=\\\"CR11\\\" class=\\\"CitationRef\\\"\\u003e2015\\u003c/span\\u003e).\\u003c/p\\u003e\\u003cp\\u003eLearning-analytics codes consistently stress transparency, purpose limitation, and so on as reported by Jisc, (\\u003cspan citationid=\\\"CR11\\\" class=\\\"CitationRef\\\"\\u003e2015\\u003c/span\\u003e), and Slade and Prinsloo, (\\u003cspan citationid=\\\"CR21\\\" class=\\\"CitationRef\\\"\\u003e2013\\u003c/span\\u003e) but in the contexts like Uganda where analytics capacity is still maturing, a synthetic-first as used in the study, an aggregate-first approach that reduces risk of inaccurate data, while institutions build policy, governance, and consent pathways. If and when approved aggregates like the anonymized LMS logins are introduced in the future, the thresholds and communications should be co-designed with students and audited for disparate impact as recommended by Slade and Prinsloo, (\\u003cspan citationid=\\\"CR21\\\" class=\\\"CitationRef\\\"\\u003e2013\\u003c/span\\u003e).\\u003c/p\\u003e\\u003cp\\u003eIn blended or online programmes, weekly structures differ but the cycles persist like the assignment deadlines, pacing and so on. However, the Prophet\\u0026rsquo;s seasonality can be re-tuned like the fortnightly cadence, and \\u0026ldquo;holiday\\u0026rdquo; regressors can encode platform-level events for the maintenance, connectivity disruptions that often generate similar dips and rebounds as supported by Hyndman and Athanasopoulos, (\\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e2021\\u003c/span\\u003e) study.\\u003c/p\\u003e\\n\\u003ch3\\u003eLimitations of the Study\\u003c/h3\\u003e\\n\\u003cp\\u003eThis study\\u0026rsquo;s benchmarks were conducted on synthetic, cohort-level series calibrated to plausible rhythms rather than observed institutional data. While this safeguards privacy and supports reproducibility, it constrains external validity and masks within-cohort heterogeneity (e.g., programme- or campus-specific patterns). Second, the purpose labels for social-media use (academic vs. recreational) are stylized and error-free in our generator, however, the real deployments will face measurement noise and shifting platform ecologies that could attenuate estimated thresholds.\\u003c/p\\u003e\\u003cp\\u003eThird, forecasts of exogenous regressors were operationalized via simple hold-forward rules of which without short-horizon models or scenario inputs for these drivers, horizon-3/4 accuracy may be overstated or understated relative to production conditions. Fourth, evaluation focused on exam-week targets and compared Prophet primarily to ARIMA in that the study did not include exponential-smoothing variants, machine-learning ensembles, or deep models, nor did not test forecast combinations, so comparative conclusions are necessarily bounded.\\u003c/p\\u003e\\u003cp\\u003eFifth, the key design choices, the holiday encodings, semester length, changepoint priors, and early-warning thresholds were tuned for the calendar like calendar idiosyncrasies, policy choices, and intervention feedback (concept drift) could alter dynamics over time, requiring periodic recalibration and fairness audits.\\u003c/p\\u003e\\u003cp\\u003eFinally, the study is predictive, not causal having a purpose-specific features that yield actionable signals which do not establish any modifying of usage alone will change outcomes without supportive, and context-aware interventions.\\u003c/p\\u003e\"},{\"header\":\"Conclusion\",\"content\":\"\\u003cp\\u003eThis study demonstrates that a calendar-aware, decomposable forecaster can transform week-to-week variability in student behavior into timely, decision-relevant signals for academic support and reporting. By explicitly modeling piecewise trends, weekly and semester seasonality, and holiday effects, Prophet consistently achieved lower forecast errors than ARIMA on the outcomes of greatest practical importance like the exam-week performance at 2\\u0026ndash;4-week horizons. This advantage aligns with Prophet\\u0026rsquo;s architectural design like the trend changepoints, multi-scale seasonality, and event regressors and with best practice in evaluation via rolling-origin cross-validation as recommended by Taylor and Letham, (\\u003cspan citationid=\\\"CR23\\\" class=\\\"CitationRef\\\"\\u003e2018\\u003c/span\\u003e), and Hyndman and Athanasopoulos, (\\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e2021\\u003c/span\\u003e). At the same time, the evidence counsel methodological humility whereby large-scale forecast competitions repeatedly show that no single method dominates across contexts, so routine benchmarking against strong statistical baselines remains essential (Makridakis, Spiliotis, \\u0026amp; Assimakopoulos, \\u003cspan citationid=\\\"CR16\\\" class=\\\"CitationRef\\\"\\u003e2018\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR17\\\" class=\\\"CitationRef\\\"\\u003e2022\\u003c/span\\u003e).\\u003c/p\\u003e\\u003cp\\u003eFor institutional adoption, an incremental, transparent, and ethical pathway is recommended from the study. Begin with cohort-level aggregates already under institutional stewardship like the weekly LMS logins or opt-in study-group activity, and so on, together with a clean holiday/recess calendar. Fit Prophet with weekly and semester seasonalities, add holidays as regressors, and include two interpretable covariates reflecting purpose (academic-oriented vs. recreational-oriented activity). Refit weekly and generate 1\\u0026ndash;4-week-ahead forecasts; evaluate with MAE and RMSE, reserving percentage errors for communication given their well-documented pathologies near zero (Hyndman \\u0026amp; Koehler, \\u003cspan citationid=\\\"CR9\\\" class=\\\"CitationRef\\\"\\u003e2006\\u003c/span\\u003e). The first implementation should be fully reproducible in which that the synthetic pipeline provided can be mirrored with institution-approved aggregates to validate feasibility prior to any consideration of finer-grained data.\\u003c/p\\u003e\\u003cp\\u003eTo ensure forecasts are actionable, not merely accurate, establish a light operating cadence, a brief weekly review in which a one-page brief presents horizon-specific forecasts, a traffic-light risk flag, and a concise note on likely drivers (e.g., recreational minutes drifting into a pre-defined caution zone). Pair alerts with enablement structured study halls, time-boxing prompts, peer-led revision sessions, or advisor outreach and focus nudges on shifting purpose rather than stigmatizing use. This implementation stance reflects the mixed empirical record on social media and learning, unstructured, heavy use correlates with lower attainment (Kirschner \\u0026amp; Karpinski, \\u003cspan citationid=\\\"CR13\\\" class=\\\"CitationRef\\\"\\u003e2010\\u003c/span\\u003e), whereas deliberate, pedagogical integration can improve engagement and grades (Junco, Heiberger, \\u0026amp; Loken, \\u003cspan citationid=\\\"CR12\\\" class=\\\"CitationRef\\\"\\u003e2011\\u003c/span\\u003e). Early-warning efforts are most effective when embedded in supportive practice rather than deployed as standalone dashboards (Arnold \\u0026amp; Pistilli, \\u003cspan citationid=\\\"CR5\\\" class=\\\"CitationRef\\\"\\u003e2012\\u003c/span\\u003e).\\u003c/p\\u003e\\u003cp\\u003eOperate aggregate-first unless and until policy, consent, and capacity justify finer granularity; document data sources, holiday encodings, modeling choices, and thresholds; audit alerts for equity across programmes and campuses; and publish a short, accessible \\u0026ldquo;how this model works\\u0026rdquo; note for students and staff (Jisc, \\u003cspan citationid=\\\"CR11\\\" class=\\\"CitationRef\\\"\\u003e2015\\u003c/span\\u003e; Slade \\u0026amp; Prinsloo, \\u003cspan citationid=\\\"CR21\\\" class=\\\"CitationRef\\\"\\u003e2013\\u003c/span\\u003e). Treat metrics with care explain MAE/RMSE to internal teams, avoid over-interpreting MAPE, and maintain rolling evaluation so models are judged on what they would have predicted at the time (Hyndman \\u0026amp; Koehler, \\u003cspan citationid=\\\"CR9\\\" class=\\\"CitationRef\\\"\\u003e2006\\u003c/span\\u003e; Hyndman \\u0026amp; Athanasopoulos, \\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e2021\\u003c/span\\u003e).\\u003c/p\\u003e\\u003cp\\u003eLooking forward, the scaffold generalizes beyond social-media features as institutions can substitute LMS engagement, library occupancy, advising traffic, or attendance and retune the semester seasonal period to local calendars while keeping outputs human-readable. As capacity grows, consider ensembles or scenario-based \\u0026ldquo;what-ifs\\u0026rdquo; (e.g., \\u0026ldquo;What if recreational scrolling declines by 20% during revision week?\\u0026rdquo;), and re-test Prophet against ARIMA/ETS and simple combinations an evidence-first discipline repeatedly rewarded in the M-competitions (Makridakis et al., \\u003cspan citationid=\\\"CR16\\\" class=\\\"CitationRef\\\"\\u003e2018\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR17\\\" class=\\\"CitationRef\\\"\\u003e2022\\u003c/span\\u003e; Hyndman \\u0026amp; Athanasopoulos, \\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e2021\\u003c/span\\u003e). With this posture of scientifically careful, ethically grounded, and student-centred, institutions can move from post-hoc diagnosis to timely, humane guidance that improves the conditions for student success (Taylor \\u0026amp; Letham, \\u003cspan citationid=\\\"CR23\\\" class=\\\"CitationRef\\\"\\u003e2018\\u003c/span\\u003e; Jisc, \\u003cspan citationid=\\\"CR11\\\" class=\\\"CitationRef\\\"\\u003e2015\\u003c/span\\u003e; Slade \\u0026amp; Prinsloo, \\u003cspan citationid=\\\"CR21\\\" class=\\\"CitationRef\\\"\\u003e2013\\u003c/span\\u003e).\\u003c/p\\u003e\"},{\"header\":\"Declarations\",\"content\":\"\\u003cp\\u003e\\u003ch2\\u003eCompeting interests:\\u003c/h2\\u003e\\u003cp\\u003eThe authors declare that they have no competing interests in this study\\u003c/p\\u003e\\u003c/p\\u003e\\u003ch2\\u003eFunding:\\u003c/h2\\u003e\\u003cp\\u003eNo funding for this study\\u003c/p\\u003e\\u003ch2\\u003eAuthor Contribution\\u003c/h2\\u003e\\u003cp\\u003eM.A. analyzed, interpreted and did the discussion section of the forecasting student performance from social-media usage. Z.N. provided the literature and did the edit writing of the manuscript. All authors read and approved the final manuscript.\\u003c/p\\u003e\\u003ch2\\u003eAvailability of data and materials:\\u003c/h2\\u003e\\u003cp\\u003eThe study used synthetic data only. No human participants were involved and no ethical approval was required\\u003c/p\\u003e\"},{\"header\":\"References\",\"content\":\"\\u003col\\u003e\\u003cli\\u003e\\u003cspan\\u003eAdewusi, M. A., Adebanjo, A. W., Odekeye, T., \\u0026amp; Kazibwe, S. (2024). Rise of the machines: Exploring the emergence of machine consciousness. \\u003cem\\u003eEuropean Journal of Theoretical and Applied Sciences\\u003c/em\\u003e, \\u003cem\\u003e2\\u003c/em\\u003e(4), 563\\u0026ndash;573.\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eAdewusi, M. A., Christine, A., Odekeye, O. T., \\u0026amp; Muhammed, N. (2025). AI in Universities: The Good, the Bot, and the Ugly Truths.\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eAdewusi, M. A., Egbowon, S. E., \\u0026amp; Akindoju, G. (2021). COVID-19 pandemic: An indigenously designed platform to the rescue. \\u003cem\\u003eJournal of Computer Science and Its Application\\u003c/em\\u003e, \\u003cem\\u003e28\\u003c/em\\u003e(2), 33\\u0026ndash;44.\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eAdewusi, M. A., Odekeye, T., Egbowon, E. S., Alade, R., \\u0026amp; Akindoju, O. G. (2022). Requirement Engineering in Learning Analytics (Machine Learning) in an Indigenously Designed Learning Platform: A Case Study. \\u003cem\\u003eAdvances in Multidisciplinary and Scientific Research Journal\\u003c/em\\u003e, \\u003cem\\u003e1\\u003c/em\\u003e, 119\\u0026ndash;126.\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eArnold, K. E., \\u0026amp; Pistilli, M. D. (2012). Course Signals at Purdue: Using learning analytics to increase student success. \\u003cem\\u003eProceedings of the 2nd International Conference on Learning Analytics and Knowledge (LAK\\u0026rsquo;12)\\u003c/em\\u003e, 267\\u0026ndash;270.\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eGSMA (2024). State of the industry report on mobile money 2024, available at: \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003e\\u003c/span\\u003e\\u003cspan address=\\\"http://www.gsma.com/sotir/\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eHsieh, M. C., Chang, M. H., Tai, Y. C., Chen, C. T., \\u0026amp; Lu, T. Y. (2025). Fast report: performance of the ETAS model in forecasting aftershock occurrence and site-specific ground-shaking intensity for the 2025 Dapu, Taiwan, earthquake sequence. \\u003cem\\u003eTerrestrial Atmospheric and Oceanic Sciences\\u003c/em\\u003e, \\u003cem\\u003e36\\u003c/em\\u003e(1), 1\\u0026ndash;8.\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eHyndman, R. J., \\u0026amp; Athanasopoulos, G. (2021). \\u003cem\\u003eForecasting: Principles and Practice\\u003c/em\\u003e (3rd ed.). OTexts. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://otexts.com/fpp3/\\u003c/span\\u003e\\u003cspan address=\\\"https://otexts.com/fpp3/\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eHyndman, R. J., \\u0026amp; Koehler, A. B. (2006). Another look at measures of forecast accuracy. \\u003cem\\u003eInternational Journal of Forecasting\\u003c/em\\u003e, \\u003cem\\u003e22\\u003c/em\\u003e(4), 679\\u0026ndash;688. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1016/j.ijforecast.2006.03.001\\u003c/span\\u003e\\u003cspan address=\\\"10.1016/j.ijforecast.2006.03.001\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eIslam, M. M., Hasan, M., Mia, M. S., Masud, A., A., \\u0026amp; Islam, A. R. M. T. (2025). \\u003cem\\u003eEarly warning systems in climate risk management: Roles and implementations in eradicating barriers and overcoming challenges\\u003c/em\\u003e. Natural Hazards Research.\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eJisc (2015). \\u003cem\\u003eCode of practice for learning analytics\\u003c/em\\u003e. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://www.jisc.ac.uk/guides/code-of-practice-for-learning-analytics\\u003c/span\\u003e\\u003cspan address=\\\"https://www.jisc.ac.uk/guides/code-of-practice-for-learning-analytics\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eJunco, R., Heiberger, G., \\u0026amp; Loken, E. (2011). The effect of Twitter on college student engagement and grades. \\u003cem\\u003eJournal of Computer Assisted Learning\\u003c/em\\u003e, \\u003cem\\u003e27\\u003c/em\\u003e(2), 119\\u0026ndash;132. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1111/j.1365-2729.2010.00387.x\\u003c/span\\u003e\\u003cspan address=\\\"10.1111/j.1365-2729.2010.00387.x\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eKirschner, P. A., \\u0026amp; Karpinski, A. C. (2010). Facebook\\u0026reg; and academic performance. \\u003cem\\u003eComputers in Human Behavior\\u003c/em\\u003e, \\u003cem\\u003e26\\u003c/em\\u003e(6), 1237\\u0026ndash;1245. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1016/j.chb.2010.03.024\\u003c/span\\u003e\\u003cspan address=\\\"10.1016/j.chb.2010.03.024\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eKuş, Z., Demirkol, S., \\u0026amp; Mert, H. (2025). Political Participation in Higher Education Just Before the 2024 Elections in Turkey: A Network Analysis. \\u003cem\\u003eDigital Society\\u003c/em\\u003e, \\u003cem\\u003e4\\u003c/em\\u003e(2), 55.\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eLyandres, E., Palazzo, B., \\u0026amp; Rabetti, D. (2022). Initial coin offering (ICO) success and post-ICO performance. \\u003cem\\u003eManagement Science\\u003c/em\\u003e, \\u003cem\\u003e68\\u003c/em\\u003e(12), 8658\\u0026ndash;8679.\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eMakridakis, S., Spiliotis, E., \\u0026amp; Assimakopoulos, V. (2018). The M4 competition: Results, findings, conclusion and way forward. \\u003cem\\u003eInternational Journal of Forecasting\\u003c/em\\u003e, \\u003cem\\u003e34\\u003c/em\\u003e(4), 802\\u0026ndash;808. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1016/j.ijforecast.2018.06.001\\u003c/span\\u003e\\u003cspan address=\\\"10.1016/j.ijforecast.2018.06.001\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eMakridakis, S., Spiliotis, E., \\u0026amp; Assimakopoulos, V. (2022). M5 accuracy competition: Results, findings, and conclusions. \\u003cem\\u003eInternational Journal of Forecasting\\u003c/em\\u003e, \\u003cem\\u003e38\\u003c/em\\u003e(4), 1346\\u0026ndash;1364. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1016/j.ijforecast.2021.11.013\\u003c/span\\u003e\\u003cspan address=\\\"10.1016/j.ijforecast.2021.11.013\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eMasalimova, A. R., Kosheleva, Y. P., Kosarenko, N. N., Kashina, S. G., Sokolova, E. G., \\u0026amp; Iakovleva, E. V. (2023). Effects of social networking sites on university students\\u0026rsquo; academic performance: A systematic review. \\u003cem\\u003eOnline Journal of Communication and Media Technologies\\u003c/em\\u003e, 13(3), e202339.\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eNIST (2018). The CSF 1.1 five functions. NIST. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://www.nist.gov/cyberframework/getting-started/online-learning/five-functions\\u003c/span\\u003e\\u003cspan address=\\\"https://www.nist.gov/cyberframework/getting-started/online-learning/five-functions\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eSalari, N., Lotfi, F., Abdolmaleki, A., Heidarian, P., Rasoulpoor, S., Fazeli, J., \\u0026amp; Mohammadi, M. (2025). The global prevalence of mild cognitive impairment in geriatric population with emphasis on influential factors: a systematic review and meta-analysis. \\u003cem\\u003eBMC geriatrics\\u003c/em\\u003e, \\u003cem\\u003e25\\u003c/em\\u003e(1), 313.\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eSlade, S., \\u0026amp; Prinsloo, P. (2013). Learning analytics: Ethical issues and dilemmas. \\u003cem\\u003eAustralasian Journal of Educational Technology\\u003c/em\\u003e, \\u003cem\\u003e29\\u003c/em\\u003e(1), 66\\u0026ndash;88. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.14742/ajet.70\\u003c/span\\u003e\\u003cspan address=\\\"10.14742/ajet.70\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eTashman, L. J. (2000). Out-of-sample tests of forecasting accuracy: an analysis and review. \\u003cem\\u003eInternational journal of forecasting\\u003c/em\\u003e, \\u003cem\\u003e16\\u003c/em\\u003e(4), 437\\u0026ndash;450.\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eTaylor, S. J., \\u0026amp; Letham, B. (2018). Forecasting at scale. \\u003cem\\u003eThe American Statistician\\u003c/em\\u003e, \\u003cem\\u003e72\\u003c/em\\u003e(1), 37\\u0026ndash;45. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1080/00031305.2017.1380080\\u003c/span\\u003e\\u003cspan address=\\\"10.1080/00031305.2017.1380080\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e\\u003c/ol\\u003e\"}],\"fulltextSource\":\"\",\"fullText\":\"\",\"funders\":[],\"hasAdminPriorityOnWorkflow\":false,\"hasManuscriptDocX\":true,\"hasOptedInToPreprint\":true,\"hasPassedJournalQc\":\"\",\"hasAnyPriority\":true,\"hideJournal\":true,\"highlight\":\"\",\"institution\":\"\",\"isAcceptedByJournal\":false,\"isAuthorSuppliedPdf\":false,\"isDeskRejected\":\"\",\"isHiddenFromSearch\":false,\"isInQc\":false,\"isInWorkflow\":false,\"isPdf\":false,\"isPdfUpToDate\":true,\"isWithdrawnOrRetracted\":false,\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"researchsquare\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":true,\"externalIdentity\":\"\",\"sideBox\":\"\",\"snPcode\":\"\",\"submissionUrl\":\"/submission\",\"title\":\"Research Square\",\"twitterHandle\":\"researchsquare\",\"acdcEnabled\":true,\"dfaEnabled\":false,\"editorialSystem\":\"\",\"reportingPortfolio\":\"\",\"inReviewEnabled\":false,\"inReviewRevisionsEnabled\":true},\"keywords\":\"Prophet model, ARIMA, Social media usage, Academic performance, Rolling cross-validation, Higher education in Uganda\",\"lastPublishedDoi\":\"10.21203/rs.3.rs-7879434/v1\",\"lastPublishedDoiUrl\":\"https://doi.org/10.21203/rs.3.rs-7879434/v1\",\"license\":{\"name\":\"CC BY 4.0\",\"url\":\"https://creativecommons.org/licenses/by/4.0/\"},\"manuscriptAbstract\":\"\\u003cp\\u003eWe evaluate time-series forecasting approaches for predicting end-of-semester examination performance using synthetic data to dynamically measure social-media usage among undergraduate students at an institution in Uganda. Building on an additive decomposition framework, Facebook\\u0026rsquo;s Prophet is specified with trend, weekly/semester seasonality, and holiday regressors reflecting national holidays and university breaks. Benchmarks include ARIMA. Model is trained with rolling cross-validation; accuracy is assessed via MAE, RMSE, and MAPE. Additionally disaggregating usage by purpose (academic vs. recreational) to test for non-linear effects and timing asymmetries around high-stakes periods. Results indicate that Prophet consistently outperforms ARIMA, particularly near changepoints (pre-exam spikes) and during holiday transitions. Purpose-specific features improve forecast accuracy and reveal thresholds were recreational use shifts from neutral to detrimental. The study further demonstrates a lightweight early-warning protocol that converts forecasts into actionable advisories for academic support units. The study advances methodological practice in Sub-Saharan higher education by moving beyond cross-sectional correlations to dynamic forecasting and offers a reproducible Python pipeline institutions to adapt for local quality-assurance and student-success initiatives.\\u003c/p\\u003e\",\"manuscriptTitle\":\"Prophet vs. ARIMA for Forecasting Student Performance from Social-Media Usage\",\"msid\":\"\",\"msnumber\":\"\",\"nonDraftVersions\":[{\"code\":1,\"date\":\"2025-10-28 16:42:57\",\"doi\":\"10.21203/rs.3.rs-7879434/v1\",\"editorialEvents\":[{\"type\":\"communityComments\",\"content\":0}],\"status\":\"published\",\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"researchsquare\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":true,\"externalIdentity\":\"\",\"sideBox\":\"\",\"snPcode\":\"\",\"submissionUrl\":\"/submission\",\"title\":\"Research Square\",\"twitterHandle\":\"researchsquare\",\"acdcEnabled\":true,\"dfaEnabled\":false,\"editorialSystem\":\"\",\"reportingPortfolio\":\"\",\"inReviewEnabled\":false,\"inReviewRevisionsEnabled\":true}}],\"origin\":\"\",\"ownerIdentity\":\"45297f59-ecf1-4383-bdfb-51bc096fc8ad\",\"owner\":[],\"postedDate\":\"October 28th, 2025\",\"published\":true,\"recentEditorialEvents\":[],\"rejectedJournal\":[],\"revision\":\"\",\"amendment\":\"\",\"status\":\"posted\",\"subjectAreas\":[],\"tags\":[],\"updatedAt\":\"2025-10-28T16:42:57+00:00\",\"versionOfRecord\":[],\"versionCreatedAt\":\"2025-10-28 16:42:57\",\"video\":\"\",\"vorDoi\":\"\",\"vorDoiUrl\":\"\",\"workflowStages\":[]},\"version\":\"v1\",\"identity\":\"rs-7879434\",\"journalConfig\":\"researchsquare\"},\"__N_SSP\":true},\"page\":\"/article/[identity]/[[...version]]\",\"query\":{\"redirect\":\"/article/rs-7879434\",\"identity\":\"rs-7879434\",\"version\":[\"v1\"]},\"buildId\":\"8U1c8b4HqxoKbykW_rLl7\",\"isFallback\":false,\"isExperimentalCompile\":false,\"dynamicIds\":[84888],\"gssp\":true,\"scriptLoader\":[]}","source_license":"CC-BY-4.0","license_restricted":false}