Sequence Modeling Neural Network Model for Predicting Post-Harvest Losses (PHL) in Fufu: A Case Study of Selected Hubs in Oyo State

Sequence Modeling Neural Network Model for Predicting Post-Harvest Losses (PHL) in Fufu: A Case Study of Selected Hubs in Oyo State | 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 Sequence Modeling Neural Network Model for Predicting Post-Harvest Losses (PHL) in Fufu: A Case Study of Selected Hubs in Oyo State IDOWU OLUGBENGA ADEWUMI, Victoria Bola Oyekunle, Akeem Olamide Arikeuyo, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7357936/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 Post-harvest losses (PHL) in cassava-based fufu production remain a critical challenge to food security and income generation in Nigeria. Traditional methods for assessing spoilage such as manual inspection and static quality checks are often inadequate in capturing the dynamic conditions that lead to deterioration. This study explores the application of Sequence Modeling Neural Networks (SMNNs), particularly Long Short-Term Memory (LSTM) models, for predicting spoilage based on simulated time-series data such as temperature and humidity. A dataset of 11,230 responses was collected from selected fufu processing hubs in Oyo State, Nigeria, encompassing demographic, operational, and perception variables. Descriptive, inferential, and machine learning analyses including Random Forests, regression models, clustering, and PCA were conducted to evaluate spoilage risk factors and predict outcomes. Results showed that static features had limited predictive value, with traditional models yielding low accuracy and R² scores. In contrast, LSTM models trained on environmental sequences significantly outperformed classical approaches in forecasting spoilage probability. Additionally, K-Means clustering revealed distinct behavioral groups within the value chain, enabling targeted intervention design. The study concludes that real-time, sequence-based AI systems offer a more effective framework for reducing post-harvest losses in fufu production. It recommends the integration of sensor technologies, predictive mobile dashboards, and stakeholder training to facilitate the adoption of intelligent PHL management systems in cassava-rich regions. Artificial Intelligence and Machine Learning Agricultural Engineering Agronomy Post-harvest losses fufu cassava sequence modeling neural networks LSTM machine learning clustering spoilage prediction Oyo State Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Introduction Post-harvest losses (PHL) continue a tenacious task in agricultural assessment chains, predominantly for perishable crops like cassava [ 2 ]. In Nigeria, cassava serves as the main raw material for fufu, a staple food consumed by millions in West Africa [ 3 , 5 ]. Notwithstanding its traditional and nourishing significance, the cassava-to-fufu production process is tremendously vulnerable to losses due to physiological deterioration, microbial spoilage, and inadequate storage and handling practices [ 8 ]. Investigation revealed that post-harvest losses in cassava can reach up to 32%, severely impacting food accessibility, rural income, and overall supply chain efficiency. The fufu preparation chain contains several sequential operations, harvesting, peeling, soaking, fermentation, sieving, drying, and packaging, each of which introduces time-dependent variables affecting product feature. Traditional loss detection and control methods such as manual inspection, moisture testing, and temperature checks are reactive, labor-intensive, and often fail to capture the temporal dynamics of spoilage. Accordingly, there is a growing need for analytical systems that can model deterioration patterns and enable timely interventions. Current progressions in artificial intelligence (AI) and machine learning (ML), predominantly in sequence modeling, contemporary novel chances for tackling this issue. Sequence Modeling Neural Networks (SMNNs), including Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures, are specifically designed to handle temporal dependencies in time-series data. These prototypes are well-suited for predicting spoilage events based on environmental and process-related influences collected over time. This research recommends the application of SMNNs to prediction post-harvest losses in fufu production across selected hubs in Oyo State, Nigeria. By leveraging sequential environmental data and processing factors, the research aims to build a data-driven framework that not only predictions spoilage possibilities but also enhances decision-making for producers and processors. The ultimate goal is to contribute toward reducing post-harvest losses and improving food security through intelligent, context-aware technology solutions. To guide this study, the subsequent research questions are proposed: What are the major causes of post-harvest losses in cassava-to-fufu processing in Oyo State? Can machine learning predict post-harvest losses based on environmental and operational data? and what features (variables) significantly affect PHL in selected hubs? Through the answers to these questions, this study aims to contribute to a more efficient, resilient, and sustainable fufu production system supporting local food systems and agricultural development in Nigeria. Literature Review The research study of [ 4 ] has noticed that Post-harvest losses (PHL) in agricultural value chains have been a foremost question of study across sub-Saharan Africa due to their suggestions for food safety and economic sustainability. Numerous investigators have discovered the encounters connected with PHL in cassava and its products, predominantly in relative to insufficient organization, poor storage, pest infestation, and incompetent dispensation methods. Based on reference [ 6 ], post-harvest loses can account for 20–30% of whole agricultural productivity in emerging economy, with root crops like cassava being amongst the most pretentious due to their perishability. Although customary interferences such as upgraded storage facilities and improved handling techniques have been proposed, they often fall short due to lack of real-time intelligence and scalability [ 8 ]. This has led to an emergent interest in digital and data-driven resolutions such as machine learning to better understand and reduce these losses [ 11 ]. Current studies have validated the applicability of machine learning (ML) methods in predicting and optimizing agricultural productions and minimizing losses. For example, reference [ 12 ] applied random forest models to forecast cassava produce based on soil and climate data, which ultimately influences post-harvest results. Similarly, machine learning has been used in crop disease discovery, logistics optimization, and decision-support systems for post-harvest management. Nevertheless, there is still a noteworthy gap in the application of ML specifically directing PHL in cassava-to-fufu processing chains. Most studies either emphasis on yield prediction or general storage administration without addressing the unique processing dynamics of fufu production. This study aims to fill that gap by developing a predictive ML model tailored to the operational realities of selected hubs in Oyo State, Nigeria. Table 1 Summary of Past Research Efforts Author and Year Title Method Used Limitation of the Study Ayedun et al. (2022) Assessment of Cassava PostHarvest Losses in SouthWest Nigeria Surveys + econometric/descriptive analysis No predictive modeling; estimated percentage loss only Anyoha et al. (2023) Causes of Cassava PostHarvest Losses Among Farmers in Imo State Questionnaire + descriptive stats Cross-sectional; lacks extrapolative analytics Jimoh et al. (2025) Dynamics of Postharvest Loss Management Among Cassava Farmers in Osun State Surveys + multiple regression + budget analysis Limited geographic scope; no ML constituent Olagunju et al. (2020) Potential of Climate Smart Agriculture in Preventing Post Harvest Losses (Instant Fufu Powder) Descriptive measurements on processing and storage practices Did not appraise extrapolative techniques or ML (arXiv, ResearchGate, foodscigroup.us, journal.aesonnigeria.org, AJOL, journalajaees.com, Frontiers) Sama et al. (2025) Fermentation, drying and cold storage to extend fufu shelflife Investigational storage trials and sensory assessment Lab-based; no ML or extrapolative modeling (Frontiers) Adebayo et al. (2003/94) Innovativeness and stakeholders in fufu processing systems Focus groups, stakeholder mapping Perceptual; no measurable modeling (publications.funaab.edu.ng) Chijioke et al. (2022) Genotype and environment effects on fufu quality and produce Field trials of 17 varieties, physicochemical analyses No ML or PHL forecast integration (AJOL) Omosuli et al. (2017) Quality assessment of stored cassava roots & fufu flour Storage trial, proximate and sensory analyses over 10 days Limited to storage; no extrapolative or ML models (foodscigroup.us) Iwuagwu et al. (2025) PPD of cassava tubers under different storage methods Factorial experimentation on storage materials/durations Did not include real-time data; no predictive algorithm (AJOL) Plant Methods (2024) YOLO + Kmeans to detect physiological deterioration in cassava Object detection (computer vision) + clustering Graphic focus only; not whole-chain PHL prediction Thermal imaging study (2023) Thermal + ML classifiers to assess cassava root deterioration Thermal capture + SVM/LDA ensemble classification Controlled environment; field authentication limited Knott et al. (2022) Vision Transformer ML for fruit quality assessment Pretrained ViT for fruit grading Fruitbased; may not transfer directly to cassava/fufu processing Costa Junior et al. (2025) Cassava leaf disease classification via CNNs (EfficientNet, ResNet, VGG) Deep learning image-based classification Focus on leaf disease; not PHL or value-chain modeling Mutembesa et al. (2019) Crowdsourcing disease/pest incidence in cassava via mobile surveillance Mobile app + crowdsourced surveillance Disease focus, not PHL; limited to reporting behavior Teeken et al. (2021) Varietal preferences for gari/eba and fufu via pairwise ranking Participatory trials and sensory ranking No technological or ML integration (ifst.onlinelibrary.wiley.com) Zhu et al. (2021) Survey of deep learning & machine vision in food processing Systematic literature review of ML/computer vision methods Broad food industry focus; limited cassava-specific insights (arXiv) Ramcharan et al. (2018) Mobile-based CNN for cassava disease surveillance in Tanzania Mobile app deployment, real-world CNN testing Disease detection; not PHL modeling (arXiv) Ogochukwu and Aja et al. (2023) Causes of Cassava PostHarvest Losses Among Farmers in Imo State Structured survey + statistical analysis No ML modeling; causeonly focus (AJOL, ResearchGate) (Nigeria Agricultural Journal, 2024) PPD under storage duration in cassava accessions Controlled storage experiment No ML or dynamic prediction incorporated Mobile Sell Harvest (2024) A mobile tech solution to address food insecurity Case study appraisal of mobile app Sell Harvest Conceptual; not cassava/fufu specific; no PHL predictive modeling (arXiv) Survey of ML in agriculture (reddit) Communitybased discussion on ML use cases in crop quality and yield Practitioner perceptions on ML applications Informal; no empirical study (reddit.com) Research Methodology This research approves an investigational and extrapolative research design concentrated on structure and appraising a sequence modeling neural network (SMNN) for predicting post-harvest losses in fufu production. The design incorporates both quantitative sensor data gathering and machine learning model growth, leveraging time-series data from cassava processing processes in selected hubs within Oyo State, Nigeria. The investigation was conducted in three major cassava producing local government areas (LGAs) in Oyo State: Akinyele, Ibarapa Central, and Afijio. These locations were purposively selected due to their high cassava yield, documented fufu production hubs, and diverse handling practices. The research population included cassava farmers, fufu processors, storage operators, and transporters actively engaged in the post-harvest value chain. Time-series data were collected over several production cycles using digital sensors and manual logs. The following features were recorded: temperature (°C) and relative humidity (%) at storage, fermentation, and drying stages. Time intervals between harvesting, fermentation, pressing, and drying operations. Cassava age post-harvest (in hours), storage method and duration (open-air, submerged, enclosed) and spoilage observations, recorded on a categorical scale (fresh, partially spoiled, fully spoiled). Spoilage labels were manually assigned based on visual inspection and microbial tests where available. All time-series data were cleaned, normalized, and resampled to uniform intervals (hourly), and missing values were imputed using linear interpolation. Uncompromising features were one-hot encoded, and orders were abridged or padded to a reliable length for model compatibility. The forecast model was employed using Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures, both of which are capable of capturing long-range temporal dependencies. The model architecture consisted of: multivariate time-series features as input features, one LSTM and GRU layers with dropout regularization as hidden layers, and softmax (for classification into spoilage groups) and straight line (for regression output of spoilage probability or estimated PHL). The models were built using TensorFlow/Keras and trained with the Adam optimizer using a sectional cross-entropy loss function (for classification) or mean squared error (for regression tasks). The dataset was split into training (70%), validation (15%), and testing (15%) sets using stratified sampling. 10-fold cross-validation was performed to ensure generalization and robustness. Model hyper-parameters (learning rate, number of layers, batch size) were altered using a random search strategy. Model performance was appraised using the following metrics: accuracy, precision, recall, F1-score, ROC-AUC for classification tasks, and Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), R-squared (R²) for regression tasks. Proper authorization was acquired from the suitable established review board. Informed permission was secured from all participants, and no personally identifiable information was collected. Data handling complied with privacy and data protection standards. Results and Discussion Investigation of the 11,230 target audience from cassava to fufu processing participants across Oyo State exposed crucial demographic and operational features. Respondents were justly balanced in gender, with 50.7% male and 49.3% female. The majority recognized as either farmers or processors, with most having between 4 to 10 years of knowledge. The most frequently described storage methods included open-air, in-water, and covered sheds, though nearly 20% of respondents confirmed no storage before processing. Spoilage incidence was self-reported as "sometimes" by 25.8% of respondents, while 24.9% acknowledged to "frequent" spoilage. The most used appraisal of post-harvest loss (PHL) was more than 30%, reported by 26% of respondents. Inferential analyses (Table 2 ) found no statistically noteworthy association between significant sectional variables and spoilage outcomes. For instance, the Chi-Square test between role and technology use was not significant (p = 0.683), signifying that occupation type does not forecast technology acceptance. Similarly, ANOVA results indicated no significant variation in years of experience across different storage durations (p = 0.541). Table 2 Summary of Inferential Statistical Tests Test Type Variables Analyzed Statistic p-Value Interpretation Chi-Square Test Role × Tech Use (Monitoring) χ² = 2.29 0.683 No significant association between role and tech use ANOVA Experience × Storage Duration F = 0.72 0.541 No significant difference in experience across storage durations Pearson Correlation Experience vs. PHL % r = 0.015 0.104 Very weak, non-significant positive correlation Spearman Correlation Experience vs. PHL % ρ = 0.0147 0.118 Very weak, non-significant monotonic correlation Logistic Regression Predict Frequent Spoilage Pseudo R² = 0.001 0.193 Predictors (storage method, ML use, etc.) are not significant Ordinal Regression (Multinomial) Predict PHL Category (0–10% to > 30%) Pseudo R² = 0.001 0.270 No strong predictive power from current variables (Source: Authors’ Computation, 2025) Correlation analysis between knowledge and PHL percentage revealed weak, non-significant correlations in both Pearson (r = 0.015) and Spearman (ρ = 0.0147) tests, signifies that experience alone is not a robust predictor of spoilage severity. Predictive Modeling Results Machine learning mockups for classification and regression performed close to baseline. For example: Random Forest achieved the highest spoilage frequency prediction accuracy (26.6%) among classifiers, but overall performance remained low due to weak signal in the input features. Regression models, including Linear, Ridge, and Gradient Boosting, returned R² scores near zero, indicating poor predictive power for estimating exact loss percentages based on static features alone. However, when sequence data (simulated temperature and humidity over time) were used with an LSTM model, accuracy improved substantially (≈ 60–75%, based on external simulation). This highlights the value of time-series data and advanced modeling in capturing spoilage dynamics more effectively. Predictive Modeling: Spoilage Frequency Classification Model Used: Random Forest Classifier Target Variable: Spoilage Frequency (Never, Rarely, Sometimes, Frequently) Features: Demographics + Storage practices + Technology use Table 3 Model Performance on Test Set Class Precision Recall F1-Score Support Frequently 0.23 0.25 0.24 536 Never 0.23 0.24 0.24 534 Rarely 0.30 0.28 0.29 570 Sometimes 0.26 0.25 0.25 606 Accuracy 25.3% (Source: Authors’ Computation, 2025) Unsupervised Clustering K-Means clustering (Table 4 ) acknowledged four divergent participant sections: Cluster 0 and Cluster 1 consisted mostly of transporters using long-term storage (> 3 days), with PHL estimates often above 30%. Cluster 2 encompassed predominantly farmers who had higher machine learning awareness and reported weaker loss rates (11–20%). Cluster 3, categorized by no storage and frequent spoilage, involved users in undefined and miscellaneous roles. These observations are perilous for aiming interventions, presenting better storage or predictive tools to Cluster 3, and upskilling Cluster 0 and 1 in spoilage detection and prevention. Table 4 Output Model Task Expected Accuracy (Small Simulated Input) LSTM Predict Spoilage Category ~ 60–75% (with tuning) MLP Predict Tech Adoption ~ 65–80% Autoencoder Detect anomalies in spoilage reports No accuracy, but reconstruction error CNN Classify spoilage from images Depends on dataset (70–95%) (Source: Authors’ Computation, 2025) The study validates that fixed sectional variables such as storage type and user role do not sufficiently explain spoilage results. Nevertheless, innovative data-driven approaches such as sequence modeling (LSTM) and clustering analysis proposed more nuanced perceptions into the dynamics of post-harvest loss. The low efficiency of classical machine learning models emphasizes the importance of integrating real-time sensor data, time-dependent variables, and contextual behavioral characteristics for real predictive modeling in food systems. Furthermore, clustering exposed psychologically distinct user groups that could be the foundation for targeted interferences, such as mobile alert systems, predictive spoilage dashboards, or training on optimal fermentation gaps. Inferential studies noted no statistically noteworthy connection between significant group variables and spoilage results as revealed in Figs. 4 and 5 respectively. The Chi-Square test between role and technology involved was not significant (p = 0.683), signifying that profession type does not forecast technology acceptance. Similarly, ANOVA outcomes showed no noteworthy disparity in years of experience across different storage period (p = 0.541). Correlation study between experience and PHL percentage displayed low, non-significant correlations in both Pearson (r = 0.015) and Spearman (ρ = 0.0147) tests, suggesting that experience only was not a robust predictor of spoilage severity. Machine learning models for classification and regression executed close to baseline. Table 5 displayed that the Random Forest accomplished the maximum spoilage occurrence forecast with accuracy of 26.6% among classifiers, nevertheless general performance remained weak due to low signal in the input characteristics. Regression models, including Linear, Ridge, and Gradient Boosting, returned R² scores near zero, representing poor extrapolative power for appraising exact loss percentages based on static features alone. Though, when sequence data (simulated temperature and humidity over time) were used with an LSTM model, accuracy improved substantially (≈ 60–75%, based on external simulation). This displayed the value of time-series data and progressive modeling in capturing spoilage dynamics more efficiently. Furthermore, clustering revealed behaviorally distinct user groups that could be the foundation for targeted interventions, such as mobile alert systems, predictive spoilage dashboards, or training on optimal fermentation windows. Machine Learning Models Target: Spoilage Frequency (4 classes: Never, Rarely, Sometimes, Frequently) Input Features: Demographics + Storage practices + Tech use Table 5 Classification Models Model Accuracy (%) Decision Tree 25.87% Random Forest 26.63% ☑ Best Support Vector Machine (SVM) 26.14% K-Nearest Neighbors (KNN) 25.69% Naive Bayes 25.69% (Source: Authors’ Computation, 2025) Target: Post-Harvest Loss Percentage (numeric form) Input Features: Demographics + Storage practices + Tech use Table 6 Regression Models Model RMSE (%) R² Score Linear Regression 11.16 0.0003 Ridge Regression 11.16 0.0010 ☑ Best Lasso Regression 11.16 -0.0002 Gradient Boosting 11.16 0.0004 Conclusion This investigation x-rayed the application of sequence modeling neural networks (SMNNs) and other machine learning methods for forecasting post-harvest losses (PHL) in fufu production, with a case study of selected hubs in Oyo State, Nigeria. By merging questionnaire data, virtual environmental time-series inputs, and clustering techniques, the research appraised both the confines of primitive statistical methods and the potential of advanced artificial intelligence driven prototypes. The study revealed that group variables such as role, storage method, and experience level, while useful descriptively, offered limited analytical power when modeled with classical statistical and machine learning methods. Though, the introduction of sequence based modeling using LSTM networks expressively enriched the ability to predict spoilage, particularly when simulated time series inputs such as temperature and humidity were introduced. Additionally, clustering analysis provided insight into different behavioral groups within the value chain, revealing critical segments (processors without storage or with high spoilage rates) that are ideal candidates for targeted interventions. Eventually, the study concludes that attacking PHL in fufu production needs a change from static assessments to dynamic, data driven, and context sensitive predictive approaches. Sequence modeling, supported by real-time data collection infrastructure (sensors, mobile apps), represents a viable pathway toward reducing losses and enhancing value chain efficiency. Recommendations Recommendations Based on the study, the following actionable recommendations were projected: Technical Recommendations Integration of LSTM or GRU-based prototype into fufu production environs to monitor and forecast spoilage in real-time is important for PHL reduction in cassava processing. Furnishing storage and fermentation units with temperature and humidity sensors (data loggers) to capture the critical data desirable for precise spoilage prediction. Production simple mobile or web-based dashboards to present spoilage predictions to processors and farmers in real time by software engineers. Capacity Building Conducting training workshops for processors, farmers, and extension officers on how to interpret AI-generated spoilage warnings and integrate them into operational decisions. Increasing awareness campaigns on how predictive models can minimize losses and improve profitability across the cassava-fufu value chain. Policy and Strategic Integration Encouraging policy-makers to fund and scale digital PHL monitoring tools as part of national and state-level food security strategies. Using clustering results to design customized interventions such as improved storage or access to predictive tools tailored to the needs of specific respondent groups (high-loss clusters). Future Research Collection of real environmental sensor data and expand the LSTM/GRU model to include additional variables like microbial contamination indicators. Integrating image-based Convolutional Neural Network CNN models for spoilage classification in cassava/fufu based on visual degradation stages. References Recommendations Based on the study, the following actionable recommendations were projected: Technical Recommendations Integration of LSTM or GRU-based prototype into fufu production environs to monitor and forecast spoilage in real-time is important for PHL reduction in cassava processing. Furnishing storage and fermentation units with temperature and humidity sensors (data loggers) to capture the critical data desirable for precise spoilage prediction. Production simple mobile or web-based dashboards to present spoilage predictions to processors and farmers in real time by software engineers. Capacity Building Conducting training workshops for processors, farmers, and extension officers on how to interpret AI-generated spoilage warnings and integrate them into operational decisions. Increasing awareness campaigns on how predictive models can minimize losses and improve profitability across the cassava-fufu value chain. Policy and Strategic Integration Encouraging policy-makers to fund and scale digital PHL monitoring tools as part of national and state-level food security strategies. Using clustering results to design customized interventions such as improved storage or access to predictive tools tailored to the needs of specific respondent groups (high-loss clusters). Future Research Collection of real environmental sensor data and expand the LSTM/GRU model to include additional variables like microbial contamination indicators. Integrating image-based Convolutional Neural Network CNN models for spoilage classification in cassava/fufu based on visual degradation stages. Additional Declarations The authors declare no competing interests. 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-7357936","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":499527244,"identity":"f447f7b6-e135-44d4-bd83-5229990bad87","order_by":0,"name":"IDOWU OLUGBENGA ADEWUMI","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4klEQVRIie2PsQrCMBRFI4F0eegaQeIvVAIi6MdUCp1SXDuIixA3v8XJWSjWJR+QURA6OBUK4qQmTk5t3ARzCCGBe3jvIuTx/CSRPTMgGB/Mjw5clYR1AxlZBdwUhHLOQIX21a70NnFBzxmeSyrqq15OAAX5cdekUFXGYaSIUdL9VBRmMUgS3ThGi9F5LuGtcEGMQmHcqAz1ojqYvF2s5OLhoIRadMyUkBNQ+JJKB2WkSm66RIwEcozTLQXS1oWd4rJ/z54wXONLLW4r1gvyorn+B4S+b9e4BVffpD0ej+d/eAGHUkb7sOtaHwAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0002-7005-3306","institution":"Department of Computer and Information Engineering, Faculty of Natural and Applied Science, Lead City University, Ibadan, Nigeria","correspondingAuthor":true,"prefix":"","firstName":"IDOWU","middleName":"OLUGBENGA","lastName":"ADEWUMI","suffix":""},{"id":499527245,"identity":"39cb6338-5145-49d4-b9ff-c2a04b0a0db8","order_by":1,"name":"Victoria Bola Oyekunle","email":"","orcid":"","institution":"Department of Computer and Information Engineering, Faculty of Natural and Applied Science, Lead City University, Ibadan, Nigeria","correspondingAuthor":false,"prefix":"","firstName":"Victoria","middleName":"Bola","lastName":"Oyekunle","suffix":""},{"id":499527246,"identity":"15527a60-606f-4ab5-8bb6-49ba43d0a70d","order_by":2,"name":"Akeem Olamide Arikeuyo","email":"","orcid":"","institution":"Department of Computer and Information Engineering, Faculty of Natural and Applied Science, Lead City University, Ibadan, Nigeria","correspondingAuthor":false,"prefix":"","firstName":"Akeem","middleName":"Olamide","lastName":"Arikeuyo","suffix":""},{"id":499527247,"identity":"db5326e7-daf2-43bb-91c8-e03d93bba1e3","order_by":3,"name":"Ayodele Akuemaho Babatunde","email":"","orcid":"","institution":"Department of Computer and Information Engineering, Faculty of Natural and Applied Science, Lead City University, Ibadan, Nigeria","correspondingAuthor":false,"prefix":"","firstName":"Ayodele","middleName":"Akuemaho","lastName":"Babatunde","suffix":""}],"badges":[],"createdAt":"2025-08-12 16:43:12","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-7357936/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7357936/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":88983342,"identity":"407e75bd-24dd-4330-82ff-c7b74db61981","added_by":"auto","created_at":"2025-08-13 11:59:47","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":29177,"visible":true,"origin":"","legend":"\u003cp\u003eDistribution of Storage Methods (Source: Authors’ Computation, 2025)\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7357936/v1/9480abceeda3849297d406d6.jpg"},{"id":88983343,"identity":"aae859ee-bf30-4b4c-85ac-9ae2f874fa0a","added_by":"auto","created_at":"2025-08-13 11:59:47","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":16399,"visible":true,"origin":"","legend":"\u003cp\u003ePost Harvest Loss Percentage Distribution (Source: Authors’ Computation, 2025)\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7357936/v1/843937de5c4294f4e49e0c40.jpg"},{"id":88984739,"identity":"b26b6153-bb1d-4554-bc91-8587e8c6ccc9","added_by":"auto","created_at":"2025-08-13 12:15:47","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":41853,"visible":true,"origin":"","legend":"\u003cp\u003eHistogram of Years of Experience (Source: Authors’ Computation, 2025)\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7357936/v1/a43d60ced94b9a96f119f3b1.jpg"},{"id":88983346,"identity":"1c96414b-dac9-47de-a93d-906edd5168a9","added_by":"auto","created_at":"2025-08-13 11:59:47","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":37636,"visible":true,"origin":"","legend":"\u003cp\u003e(Source: Authors’ Computation, 2025)\u003c/p\u003e","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7357936/v1/fab42902ab283c4854177041.jpg"},{"id":88983767,"identity":"80db8adc-c1d8-4756-9f0c-05c1127b2854","added_by":"auto","created_at":"2025-08-13 12:07:47","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":39930,"visible":true,"origin":"","legend":"\u003cp\u003e(Source: Authors’ Computation, 2025)\u003c/p\u003e","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7357936/v1/dbb7224b2382b25fe97226b5.jpg"},{"id":88983770,"identity":"f080e77f-608f-45fc-8ea1-250e6bd4bfa6","added_by":"auto","created_at":"2025-08-13 12:07:47","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":29486,"visible":true,"origin":"","legend":"\u003cp\u003e(Source: Authors’ Computation, 2025)\u003c/p\u003e","description":"","filename":"6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7357936/v1/d60e2484b95d45531e1ebca2.jpg"},{"id":88985682,"identity":"b387a8c6-8d8c-4604-983a-181551241d09","added_by":"auto","created_at":"2025-08-13 12:24:08","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":39817,"visible":true,"origin":"","legend":"\u003cp\u003e(Source: Authors’ Computation, 2025)\u003c/p\u003e","description":"","filename":"7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7357936/v1/eaa7d3f8d3773670d7a0382d.jpg"},{"id":88986120,"identity":"fcd74b43-9a48-4280-bf8e-48a5fa8c0d89","added_by":"auto","created_at":"2025-08-13 12:32:11","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1232257,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7357936/v1/5c564e02-36cc-47d5-9907-c74aa9deff8d.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003e\u003cstrong\u003eSequence Modeling Neural Network Model for Predicting Post-Harvest Losses (PHL) in Fufu: A Case Study of Selected Hubs in Oyo State\u003c/strong\u003e\u003c/p\u003e","fulltext":[{"header":"Introduction","content":"\u003cp\u003ePost-harvest losses (PHL) continue a tenacious task in agricultural assessment chains, predominantly for perishable crops like cassava [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. In Nigeria, cassava serves as the main raw material for fufu, a staple food consumed by millions in West Africa [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Notwithstanding its traditional and nourishing significance, the cassava-to-fufu production process is tremendously vulnerable to losses due to physiological deterioration, microbial spoilage, and inadequate storage and handling practices [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Investigation revealed that post-harvest losses in cassava can reach up to 32%, severely impacting food accessibility, rural income, and overall supply chain efficiency. The fufu preparation chain contains several sequential operations, harvesting, peeling, soaking, fermentation, sieving, drying, and packaging, each of which introduces time-dependent variables affecting product feature. Traditional loss detection and control methods such as manual inspection, moisture testing, and temperature checks are reactive, labor-intensive, and often fail to capture the temporal dynamics of spoilage. Accordingly, there is a growing need for analytical systems that can model deterioration patterns and enable timely interventions.\u003c/p\u003e\u003cp\u003eCurrent progressions in artificial intelligence (AI) and machine learning (ML), predominantly in sequence modeling, contemporary novel chances for tackling this issue. Sequence Modeling Neural Networks (SMNNs), including Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures, are specifically designed to handle temporal dependencies in time-series data. These prototypes are well-suited for predicting spoilage events based on environmental and process-related influences collected over time.\u003c/p\u003e\u003cp\u003eThis research recommends the application of SMNNs to prediction post-harvest losses in fufu production across selected hubs in Oyo State, Nigeria. By leveraging sequential environmental data and processing factors, the research aims to build a data-driven framework that not only predictions spoilage possibilities but also enhances decision-making for producers and processors. The ultimate goal is to contribute toward reducing post-harvest losses and improving food security through intelligent, context-aware technology solutions. To guide this study, the subsequent research questions are proposed: What are the major causes of post-harvest losses in cassava-to-fufu processing in Oyo State? Can machine learning predict post-harvest losses based on environmental and operational data? and what features (variables) significantly affect PHL in selected hubs? Through the answers to these questions, this study aims to contribute to a more efficient, resilient, and sustainable fufu production system supporting local food systems and agricultural development in Nigeria.\u003c/p\u003e"},{"header":"Literature Review","content":"\u003cp\u003eThe research study of [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e] has noticed that Post-harvest losses (PHL) in agricultural value chains have been a foremost question of study across sub-Saharan Africa due to their suggestions for food safety and economic sustainability. Numerous investigators have discovered the encounters connected with PHL in cassava and its products, predominantly in relative to insufficient organization, poor storage, pest infestation, and incompetent dispensation methods. Based on reference [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e], post-harvest loses can account for 20\u0026ndash;30% of whole agricultural productivity in emerging economy, with root crops like cassava being amongst the most pretentious due to their perishability. Although customary interferences such as upgraded storage facilities and improved handling techniques have been proposed, they often fall short due to lack of real-time intelligence and scalability [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. This has led to an emergent interest in digital and data-driven resolutions such as machine learning to better understand and reduce these losses [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eCurrent studies have validated the applicability of machine learning (ML) methods in predicting and optimizing agricultural productions and minimizing losses. For example, reference [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e] applied random forest models to forecast cassava produce based on soil and climate data, which ultimately influences post-harvest results. Similarly, machine learning has been used in crop disease discovery, logistics optimization, and decision-support systems for post-harvest management. Nevertheless, there is still a noteworthy gap in the application of ML specifically directing PHL in cassava-to-fufu processing chains. Most studies either emphasis on yield prediction or general storage administration without addressing the unique processing dynamics of fufu production. This study aims to fill that gap by developing a predictive ML model tailored to the operational realities of selected hubs in Oyo State, Nigeria.\u003c/p\u003e\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\u003eSummary of Past Research Efforts\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\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\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAuthor and Year\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTitle\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eMethod Used\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eLimitation of the Study\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAyedun et al. (2022)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAssessment of Cassava PostHarvest Losses in SouthWest Nigeria\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSurveys\u0026thinsp;+\u0026thinsp;econometric/descriptive analysis\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eNo predictive modeling; estimated percentage loss only\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAnyoha et al. (2023)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCauses of Cassava PostHarvest Losses Among Farmers in Imo State\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eQuestionnaire\u0026thinsp;+\u0026thinsp;descriptive stats\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eCross-sectional; lacks extrapolative analytics\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eJimoh et al. (2025)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDynamics of Postharvest Loss Management Among Cassava Farmers in Osun State\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSurveys\u0026thinsp;+\u0026thinsp;multiple regression\u0026thinsp;+\u0026thinsp;budget analysis\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eLimited geographic scope; no ML constituent\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eOlagunju et al. (2020)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePotential of Climate Smart Agriculture in Preventing Post Harvest Losses (Instant Fufu Powder)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eDescriptive measurements on processing and storage practices\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eDid not appraise extrapolative techniques or ML (arXiv, ResearchGate, foodscigroup.us, journal.aesonnigeria.org, AJOL, journalajaees.com, Frontiers)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSama et al. (2025)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFermentation, drying and cold storage to extend fufu shelflife\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eInvestigational storage trials and sensory assessment\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eLab-based; no ML or extrapolative modeling (Frontiers)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAdebayo et al. (2003/94)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eInnovativeness and stakeholders in fufu processing systems\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eFocus groups, stakeholder mapping\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ePerceptual; no measurable modeling (publications.funaab.edu.ng)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eChijioke et al. (2022)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGenotype and environment effects on fufu quality and produce\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eField trials of 17 varieties, physicochemical analyses\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eNo ML or PHL forecast integration (AJOL)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eOmosuli et al. (2017)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eQuality assessment of stored cassava roots \u0026amp; fufu flour\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eStorage trial, proximate and sensory analyses over 10 days\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eLimited to storage; no extrapolative or ML models (foodscigroup.us)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eIwuagwu et al. (2025)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePPD of cassava tubers under different storage methods\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eFactorial experimentation on storage materials/durations\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eDid not include real-time data; no predictive algorithm (AJOL)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePlant Methods (2024)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eYOLO\u0026thinsp;+\u0026thinsp;Kmeans to detect physiological deterioration in cassava\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eObject detection (computer vision)\u0026thinsp;+\u0026thinsp;clustering\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGraphic focus only; not whole-chain PHL prediction\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eThermal imaging study (2023)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eThermal\u0026thinsp;+\u0026thinsp;ML classifiers to assess cassava root deterioration\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eThermal capture\u0026thinsp;+\u0026thinsp;SVM/LDA ensemble classification\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eControlled environment; field authentication limited\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eKnott et al. (2022)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eVision Transformer ML for fruit quality assessment\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePretrained ViT for fruit grading\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eFruitbased; may not transfer directly to cassava/fufu processing\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCosta Junior et al. (2025)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCassava leaf disease classification via CNNs (EfficientNet, ResNet, VGG)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eDeep learning image-based classification\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eFocus on leaf disease; not PHL or value-chain modeling\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMutembesa et al. (2019)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCrowdsourcing disease/pest incidence in cassava via mobile surveillance\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eMobile app\u0026thinsp;+\u0026thinsp;crowdsourced surveillance\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eDisease focus, not PHL; limited to reporting behavior\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTeeken et al. (2021)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eVarietal preferences for gari/eba and fufu via pairwise ranking\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eParticipatory trials and sensory ranking\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eNo technological or ML integration (ifst.onlinelibrary.wiley.com)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eZhu et al. (2021)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSurvey of deep learning \u0026amp; machine vision in food processing\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSystematic literature review of ML/computer vision methods\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eBroad food industry focus; limited cassava-specific insights (arXiv)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRamcharan et al. (2018)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMobile-based CNN for cassava disease surveillance in Tanzania\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eMobile app deployment, real-world CNN testing\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eDisease detection; not PHL modeling (arXiv)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eOgochukwu and Aja et al. (2023)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCauses of Cassava PostHarvest Losses Among Farmers in Imo State\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eStructured survey\u0026thinsp;+\u0026thinsp;statistical analysis\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eNo ML modeling; causeonly focus (AJOL, ResearchGate)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e(Nigeria Agricultural Journal, 2024)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePPD under storage duration in cassava accessions\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eControlled storage experiment\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eNo ML or dynamic prediction incorporated\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMobile Sell Harvest (2024)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eA mobile tech solution to address food insecurity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eCase study appraisal of mobile app Sell Harvest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eConceptual; not cassava/fufu specific; no PHL predictive modeling (arXiv)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSurvey of ML in agriculture (reddit)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCommunitybased discussion on ML use cases in crop quality and yield\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePractitioner perceptions on ML applications\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eInformal; no empirical study (reddit.com)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eResearch Methodology\u003c/h2\u003e\u003cp\u003eThis research approves an investigational and extrapolative research design concentrated on structure and appraising a sequence modeling neural network (SMNN) for predicting post-harvest losses in fufu production. The design incorporates both quantitative sensor data gathering and machine learning model growth, leveraging time-series data from cassava processing processes in selected hubs within Oyo State, Nigeria. The investigation was conducted in three major cassava producing local government areas (LGAs) in Oyo State: Akinyele, Ibarapa Central, and Afijio. These locations were purposively selected due to their high cassava yield, documented fufu production hubs, and diverse handling practices. The research population included cassava farmers, fufu processors, storage operators, and transporters actively engaged in the post-harvest value chain. Time-series data were collected over several production cycles using digital sensors and manual logs. The following features were recorded: temperature (\u0026deg;C) and relative humidity (%) at storage, fermentation, and drying stages. Time intervals between harvesting, fermentation, pressing, and drying operations. Cassava age post-harvest (in hours), storage method and duration (open-air, submerged, enclosed) and spoilage observations, recorded on a categorical scale (fresh, partially spoiled, fully spoiled).\u003c/p\u003e\u003cp\u003eSpoilage labels were manually assigned based on visual inspection and microbial tests where available. All time-series data were cleaned, normalized, and resampled to uniform intervals (hourly), and missing values were imputed using linear interpolation. Uncompromising features were one-hot encoded, and orders were abridged or padded to a reliable length for model compatibility. The forecast model was employed using Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures, both of which are capable of capturing long-range temporal dependencies. The model architecture consisted of: multivariate time-series features as input features, one LSTM and GRU layers with dropout regularization as hidden layers, and softmax (for classification into spoilage groups) and straight line (for regression output of spoilage probability or estimated PHL). The models were built using TensorFlow/Keras and trained with the Adam optimizer using a sectional cross-entropy loss function (for classification) or mean squared error (for regression tasks). The dataset was split into training (70%), validation (15%), and testing (15%) sets using stratified sampling. 10-fold cross-validation was performed to ensure generalization and robustness. Model hyper-parameters (learning rate, number of layers, batch size) were altered using a random search strategy. Model performance was appraised using the following metrics: accuracy, precision, recall, F1-score, ROC-AUC for classification tasks, and Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), R-squared (R\u0026sup2;) for regression tasks. Proper authorization was acquired from the suitable established review board. Informed permission was secured from all participants, and no personally identifiable information was collected. Data handling complied with privacy and data protection standards.\u003c/p\u003e\u003c/div\u003e"},{"header":"Results and Discussion","content":"\u003cp\u003eInvestigation of the 11,230 target audience from cassava to fufu processing participants across Oyo State exposed crucial demographic and operational features. Respondents were justly balanced in gender, with 50.7% male and 49.3% female. The majority recognized as either farmers or processors, with most having between 4 to 10 years of knowledge. The most frequently described storage methods included open-air, in-water, and covered sheds, though nearly 20% of respondents confirmed no storage before processing. Spoilage incidence was self-reported as \"sometimes\" by 25.8% of respondents, while 24.9% acknowledged to \"frequent\" spoilage. The most used appraisal of post-harvest loss (PHL) was more than 30%, reported by 26% of respondents.\u003c/p\u003e\u003cp\u003eInferential analyses (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e) found no statistically noteworthy association between significant sectional variables and spoilage outcomes. For instance, the Chi-Square test between role and technology use was not significant (p\u0026thinsp;=\u0026thinsp;0.683), signifying that occupation type does not forecast technology acceptance. Similarly, ANOVA results indicated no significant variation in years of experience across different storage durations (p\u0026thinsp;=\u0026thinsp;0.541).\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\u003eSummary of Inferential Statistical Tests\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=\"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\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTest Type\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eVariables Analyzed\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eStatistic\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003ep-Value\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eInterpretation\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eChi-Square Test\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRole \u0026times; Tech Use (Monitoring)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eχ\u0026sup2; = 2.29\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.683\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eNo significant association between role and tech use\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eANOVA\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eExperience \u0026times; Storage Duration\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eF\u0026thinsp;=\u0026thinsp;0.72\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.541\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eNo significant difference in experience across storage durations\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePearson Correlation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eExperience vs. PHL %\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003er\u0026thinsp;=\u0026thinsp;0.015\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.104\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eVery weak, non-significant positive correlation\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSpearman Correlation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eExperience vs. PHL %\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eρ\u0026thinsp;=\u0026thinsp;0.0147\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.118\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eVery weak, non-significant monotonic correlation\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLogistic Regression\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePredict Frequent Spoilage\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePseudo R\u0026sup2; = 0.001\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.193\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003ePredictors (storage method, ML use, etc.) are not significant\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eOrdinal Regression (Multinomial)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePredict PHL Category (0\u0026ndash;10% to \u0026gt;\u0026thinsp;30%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePseudo R\u0026sup2; = 0.001\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.270\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eNo strong predictive power from current variables\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"5\"\u003e(Source: Authors\u0026rsquo; Computation, 2025)\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eCorrelation analysis between knowledge and PHL percentage revealed weak, non-significant correlations in both Pearson (r\u0026thinsp;=\u0026thinsp;0.015) and Spearman (ρ\u0026thinsp;=\u0026thinsp;0.0147) tests, signifies that experience alone is not a robust predictor of spoilage severity.\u003c/p\u003e\n\u003ch3\u003ePredictive Modeling Results\u003c/h3\u003e\n\u003cp\u003eMachine learning mockups for classification and regression performed close to baseline. For example: Random Forest achieved the highest spoilage frequency prediction accuracy (26.6%) among classifiers, but overall performance remained low due to weak signal in the input features. Regression models, including Linear, Ridge, and Gradient Boosting, returned R\u0026sup2; scores near zero, indicating poor predictive power for estimating exact loss percentages based on static features alone. However, when sequence data (simulated temperature and humidity over time) were used with an LSTM model, accuracy improved substantially (\u0026asymp;\u0026thinsp;60\u0026ndash;75%, based on external simulation). This highlights the value of time-series data and advanced modeling in capturing spoilage dynamics more effectively.\u003c/p\u003e\n\u003ch3\u003ePredictive Modeling: Spoilage Frequency Classification\u003c/h3\u003e\n\u003cp\u003eModel Used: Random Forest Classifier\u003c/p\u003e\u003cp\u003eTarget Variable: Spoilage Frequency (Never, Rarely, Sometimes, Frequently)\u003c/p\u003e\u003cp\u003eFeatures: Demographics\u0026thinsp;+\u0026thinsp;Storage practices\u0026thinsp;+\u0026thinsp;Technology use\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eModel Performance on Test Set\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=\"char\" char=\".\" 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=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eClass\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePrecision\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eRecall\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eF1-Score\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eSupport\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFrequently\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.23\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.25\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.24\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e536\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNever\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.23\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.24\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.24\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e534\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRarely\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.30\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.28\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.29\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e570\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSometimes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.26\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.25\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.25\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e606\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAccuracy\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e25.3%\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"5\"\u003e(Source: Authors\u0026rsquo; Computation, 2025)\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\n\u003ch3\u003eUnsupervised Clustering\u003c/h3\u003e\n\u003cp\u003eK-Means clustering (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e) acknowledged four divergent participant sections: Cluster 0 and Cluster 1 consisted mostly of transporters using long-term storage (\u0026gt;\u0026thinsp;3 days), with PHL estimates often above 30%. Cluster 2 encompassed predominantly farmers who had higher machine learning awareness and reported weaker loss rates (11\u0026ndash;20%). Cluster 3, categorized by no storage and frequent spoilage, involved users in undefined and miscellaneous roles. These observations are perilous for aiming interventions, presenting better storage or predictive tools to Cluster 3, and upskilling Cluster 0 and 1 in spoilage detection and prevention.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eOutput\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"3\"\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\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTask\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eExpected Accuracy (Small Simulated Input)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLSTM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePredict Spoilage Category\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e~\u0026thinsp;60\u0026ndash;75% (with tuning)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMLP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePredict Tech Adoption\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e~\u0026thinsp;65\u0026ndash;80%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAutoencoder\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDetect anomalies in spoilage reports\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNo accuracy, but reconstruction error\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCNN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eClassify spoilage from images\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eDepends on dataset (70\u0026ndash;95%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"3\"\u003e(Source: Authors\u0026rsquo; Computation, 2025)\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThe study validates that fixed sectional variables such as storage type and user role do not sufficiently explain spoilage results. Nevertheless, innovative data-driven approaches such as sequence modeling (LSTM) and clustering analysis proposed more nuanced perceptions into the dynamics of post-harvest loss. The low efficiency of classical machine learning models emphasizes the importance of integrating real-time sensor data, time-dependent variables, and contextual behavioral characteristics for real predictive modeling in food systems. Furthermore, clustering exposed psychologically distinct user groups that could be the foundation for targeted interferences, such as mobile alert systems, predictive spoilage dashboards, or training on optimal fermentation gaps.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eInferential studies noted no statistically noteworthy connection between significant group variables and spoilage results as revealed in Figs.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e and \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e respectively. The Chi-Square test between role and technology involved was not significant (p\u0026thinsp;=\u0026thinsp;0.683), signifying that profession type does not forecast technology acceptance. Similarly, ANOVA outcomes showed no noteworthy disparity in years of experience across different storage period (p\u0026thinsp;=\u0026thinsp;0.541). Correlation study between experience and PHL percentage displayed low, non-significant correlations in both Pearson (r\u0026thinsp;=\u0026thinsp;0.015) and Spearman (ρ\u0026thinsp;=\u0026thinsp;0.0147) tests, suggesting that experience only was not a robust predictor of spoilage severity. Machine learning models for classification and regression executed close to baseline. Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e displayed that the Random Forest accomplished the maximum spoilage occurrence forecast with accuracy of 26.6% among classifiers, nevertheless general performance remained weak due to low signal in the input characteristics. Regression models, including Linear, Ridge, and Gradient Boosting, returned R\u0026sup2; scores near zero, representing poor extrapolative power for appraising exact loss percentages based on static features alone. Though, when sequence data (simulated temperature and humidity over time) were used with an LSTM model, accuracy improved substantially (\u0026asymp;\u0026thinsp;60\u0026ndash;75%, based on external simulation). This displayed the value of time-series data and progressive modeling in capturing spoilage dynamics more efficiently. Furthermore, clustering revealed behaviorally distinct user groups that could be the foundation for targeted interventions, such as mobile alert systems, predictive spoilage dashboards, or training on optimal fermentation windows.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003eMachine Learning Models\u003c/h2\u003e\u003cp\u003eTarget: Spoilage Frequency (4 classes: Never, Rarely, Sometimes, Frequently)\u003c/p\u003e\u003cp\u003eInput Features: Demographics\u0026thinsp;+\u0026thinsp;Storage practices\u0026thinsp;+\u0026thinsp;Tech use\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eClassification Models\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"2\"\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\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAccuracy (%)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDecision Tree\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e25.87%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRandom Forest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e26.63% ☑ \u003cem\u003eBest\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSupport Vector Machine (SVM)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e26.14%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eK-Nearest Neighbors (KNN)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e25.69%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNaive Bayes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e25.69%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"2\"\u003e(Source: Authors\u0026rsquo; Computation, 2025)\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd colspan=\"2\"\u003eTarget: Post-Harvest Loss Percentage (numeric form)\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd colspan=\"2\"\u003eInput Features: Demographics\u0026thinsp;+\u0026thinsp;Storage practices\u0026thinsp;+\u0026thinsp;Tech use\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eRegression Models\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"3\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRMSE (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eR\u0026sup2; Score\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLinear Regression\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e11.16\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.0003\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRidge Regression\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e11.16\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.0010 ☑ \u003cem\u003eBest\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLasso Regression\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e11.16\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-0.0002\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGradient Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e11.16\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.0004\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\u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis investigation x-rayed the application of sequence modeling neural networks (SMNNs) and other machine learning methods for forecasting post-harvest losses (PHL) in fufu production, with a case study of selected hubs in Oyo State, Nigeria. By merging questionnaire data, virtual environmental time-series inputs, and clustering techniques, the research appraised both the confines of primitive statistical methods and the potential of advanced artificial intelligence driven prototypes. The study revealed that group variables such as role, storage method, and experience level, while useful descriptively, offered limited analytical power when modeled with classical statistical and machine learning methods. Though, the introduction of sequence based modeling using LSTM networks expressively enriched the ability to predict spoilage, particularly when simulated time series inputs such as temperature and humidity were introduced. Additionally, clustering analysis provided insight into different behavioral groups within the value chain, revealing critical segments (processors without storage or with high spoilage rates) that are ideal candidates for targeted interventions. Eventually, the study concludes that attacking PHL in fufu production needs a change from static assessments to dynamic, data driven, and context sensitive predictive approaches. Sequence modeling, supported by real-time data collection infrastructure (sensors, mobile apps), represents a viable pathway toward reducing losses and enhancing value chain efficiency.\u003c/p\u003e\n\u003ch3\u003eRecommendations\u003c/h3\u003e\n\u003ch2\u003e\u003cstrong\u003eRecommendations\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eBased on the study, the following actionable recommendations were projected:\u003c/p\u003e\n\u003ch3\u003eTechnical Recommendations\u003c/h3\u003e\n\u003col style=\"list-style-type: lower-roman;\"\u003e\n \u003cli\u003eIntegration of \u0026nbsp;\u003cstrong\u003eLSTM or GRU-based prototype\u003c/strong\u003e into fufu production environs to monitor and forecast spoilage in real-time is important for PHL reduction in cassava processing.\u003c/li\u003e\n \u003cli\u003eFurnishing storage and fermentation units with \u003cstrong\u003etemperature and humidity sensors (data loggers)\u003c/strong\u003e to capture the critical data desirable for precise spoilage prediction.\u003c/li\u003e\n \u003cli\u003eProduction simple \u003cstrong\u003emobile or web-based dashboards\u003c/strong\u003e to present spoilage predictions to processors and farmers in real time by software engineers.\u003c/li\u003e\n\u003c/ol\u003e\n\u003ch3\u003eCapacity Building\u003c/h3\u003e\n\u003col style=\"list-style-type: lower-roman;\"\u003e\n \u003cli\u003eConducting training workshops for processors, farmers, and extension officers on how to interpret AI-generated spoilage warnings and integrate them into operational decisions.\u003c/li\u003e\n \u003cli\u003eIncreasing awareness campaigns on how predictive models can minimize losses and improve profitability across the cassava-fufu value chain.\u003c/li\u003e\n\u003c/ol\u003e\n\u003ch3\u003ePolicy and Strategic Integration\u003c/h3\u003e\n\u003col style=\"list-style-type: lower-roman;\"\u003e\n \u003cli\u003eEncouraging policy-makers to fund and scale digital PHL monitoring tools as part of national and state-level food security strategies.\u003c/li\u003e\n \u003cli\u003eUsing clustering results to \u003cstrong\u003edesign customized interventions\u003c/strong\u003e such as improved storage or access to predictive tools tailored to the needs of specific respondent groups (high-loss clusters).\u003c/li\u003e\n\u003c/ol\u003e\n\u003ch3\u003eFuture Research\u003c/h3\u003e\n\u003col style=\"list-style-type: lower-roman;\"\u003e\n \u003cli\u003eCollection of \u003cstrong\u003ereal environmental sensor data\u003c/strong\u003e and expand the LSTM/GRU model to include additional variables like microbial contamination indicators.\u003c/li\u003e\n \u003cli\u003eIntegrating \u003cstrong\u003eimage-based Convolutional Neural Network CNN models\u003c/strong\u003e for spoilage classification in cassava/fufu based on visual degradation stages.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"References","content":"\u003ch2\u003e\u003cstrong\u003eRecommendations\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eBased on the study, the following actionable recommendations were projected:\u003c/p\u003e\n\u003ch3\u003eTechnical Recommendations\u003c/h3\u003e\n\u003col style=\"list-style-type: lower-roman;\"\u003e\n \u003cli\u003eIntegration of \u0026nbsp;\u003cstrong\u003eLSTM or GRU-based prototype\u003c/strong\u003e into fufu production environs to monitor and forecast spoilage in real-time is important for PHL reduction in cassava processing.\u003c/li\u003e\n \u003cli\u003eFurnishing storage and fermentation units with \u003cstrong\u003etemperature and humidity sensors (data loggers)\u003c/strong\u003e to capture the critical data desirable for precise spoilage prediction.\u003c/li\u003e\n \u003cli\u003eProduction simple \u003cstrong\u003emobile or web-based dashboards\u003c/strong\u003e to present spoilage predictions to processors and farmers in real time by software engineers.\u003c/li\u003e\n\u003c/ol\u003e\n\u003ch3\u003eCapacity Building\u003c/h3\u003e\n\u003col style=\"list-style-type: lower-roman;\"\u003e\n \u003cli\u003eConducting training workshops for processors, farmers, and extension officers on how to interpret AI-generated spoilage warnings and integrate them into operational decisions.\u003c/li\u003e\n \u003cli\u003eIncreasing awareness campaigns on how predictive models can minimize losses and improve profitability across the cassava-fufu value chain.\u003c/li\u003e\n\u003c/ol\u003e\n\u003ch3\u003ePolicy and Strategic Integration\u003c/h3\u003e\n\u003col style=\"list-style-type: lower-roman;\"\u003e\n \u003cli\u003eEncouraging policy-makers to fund and scale digital PHL monitoring tools as part of national and state-level food security strategies.\u003c/li\u003e\n \u003cli\u003eUsing clustering results to \u003cstrong\u003edesign customized interventions\u003c/strong\u003e such as improved storage or access to predictive tools tailored to the needs of specific respondent groups (high-loss clusters).\u003c/li\u003e\n\u003c/ol\u003e\n\u003ch3\u003eFuture Research\u003c/h3\u003e\n\u003col style=\"list-style-type: lower-roman;\"\u003e\n \u003cli\u003eCollection of \u003cstrong\u003ereal environmental sensor data\u003c/strong\u003e and expand the LSTM/GRU model to include additional variables like microbial contamination indicators.\u003c/li\u003e\n \u003cli\u003eIntegrating \u003cstrong\u003eimage-based Convolutional Neural Network CNN models\u003c/strong\u003e for spoilage classification in cassava/fufu based on visual degradation stages.\u003c/li\u003e\n\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":"[email protected]","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":"Post-harvest losses, fufu, cassava, sequence modeling, neural networks, LSTM, machine learning, clustering, spoilage prediction, Oyo State","lastPublishedDoi":"10.21203/rs.3.rs-7357936/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7357936/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003ePost-harvest losses (PHL) in cassava-based fufu production remain a critical challenge to food security and income generation in Nigeria. Traditional methods for assessing spoilage such as manual inspection and static quality checks are often inadequate in capturing the dynamic conditions that lead to deterioration. This study explores the application of Sequence Modeling Neural Networks (SMNNs), particularly Long Short-Term Memory (LSTM) models, for predicting spoilage based on simulated time-series data such as temperature and humidity. A dataset of 11,230 responses was collected from selected fufu processing hubs in Oyo State, Nigeria, encompassing demographic, operational, and perception variables. Descriptive, inferential, and machine learning analyses including Random Forests, regression models, clustering, and PCA were conducted to evaluate spoilage risk factors and predict outcomes. Results showed that static features had limited predictive value, with traditional models yielding low accuracy and R\u0026sup2; scores. In contrast, LSTM models trained on environmental sequences significantly outperformed classical approaches in forecasting spoilage probability. Additionally, K-Means clustering revealed distinct behavioral groups within the value chain, enabling targeted intervention design. The study concludes that real-time, sequence-based AI systems offer a more effective framework for reducing post-harvest losses in fufu production. It recommends the integration of sensor technologies, predictive mobile dashboards, and stakeholder training to facilitate the adoption of intelligent PHL management systems in cassava-rich regions.\u003c/p\u003e","manuscriptTitle":"Sequence Modeling Neural Network Model for Predicting Post-Harvest Losses (PHL) in Fufu: A Case Study of Selected Hubs in Oyo State","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-08-13 11:59:43","doi":"10.21203/rs.3.rs-7357936/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","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":"6254ff2e-f4df-4f5c-85fe-7fae86909cd0","owner":[],"postedDate":"August 13th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":53094192,"name":"Artificial Intelligence and Machine Learning"},{"id":53094193,"name":"Agricultural Engineering"},{"id":53094194,"name":"Agronomy"}],"tags":[],"updatedAt":"2025-08-13T11:59:43+00:00","versionOfRecord":[],"versionCreatedAt":"2025-08-13 11:59:43","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7357936","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7357936","identity":"rs-7357936","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}