Pesticide Poisoning Diagnosis System

Abstract

Pesticideexposureisacriticalpublichealthissueinagriculturalcommunities, especiallyamongruralworkerswithlimited access to timely medical care. Inaccurate or delayed diagnosis due to vague symptom patterns often leads to severe health outcomes. This research proposes an AI-based diagnostic system that leverages an ensemble of supervised machine learning algorithms—K- Nearest Neighbors, Logistic Regression, Gradient Boosting, and Random Forest—to predict pesticide poisoning cases with high accuracy. The model processes user-provided data, such as symptom inputs and exposure details, and combines predictions through majority voting to enhance reliability. A user-friendly web interface enables real-time diagnosis and result visualization, ensuring accessibility for non-technical users in rural areas. Experimental evaluation shows that the ensemble system achieves over 96% accuracy, with robust performance in both precision and recall. The proposed approach offers a scalable, data-driven solution to bridge the healthcare gap in underserved regions and support early medical intervention for pesticide-related illnesses.

Introduction

Pesticide poisoning poses a serious health risk for rural agricultural workers, who often face limited medical access and delayed diagnosis. Traditional diagnostic methods are manual, inconsistent, and inadequate for rural settings, especially given overlapping symptoms and unstructured data. To address these challenges, this project implements an AI-driven, ensemble-based diagnostic system combining multiple machine learning models—K-Nearest Neighbors, Logistic Regression, Gradient Boosting, and Random Forest—to improve accuracy in predicting pesticide poisoning from user-reported symptoms.

The system uses a structured, synthetic dataset augmented with real-world poisoning data, processed through standard data preparation steps. A web-based interface built with Django allows real-time symptom input and instant diagnosis display, featuring user-friendly forms and graphical insights to aid understanding. The ensemble approach leverages soft voting to combine individual model outputs, enhancing predictive performance and robustness in rural healthcare contexts.

Evaluation metrics (accuracy, precision, recall, F1-score) confirm the system’s effectiveness in delivering rapid, reliable diagnoses, potentially bridging critical gaps in early medical intervention for under-resourced communities. The platform emphasizes accessibility, data security, and usability to encourage adoption and improve health outcomes in rural areas.

Conclusion

This research demonstrates that an ensemble-based diagnostic system, when trained on a hybrid dataset combining clinical, environmental,andAI-generateddata,achieveshighaccuracy(F1-score:0.97,accuracy:96%)inpredictingpesticidepoisoning cases among rural populations. The proposed framework effectively addresses the diagnostic challenges faced in underserved regions by: A. Multi-ModelEnsembleLearning: IntegratingKNN,LogisticRegression,RandomForest,andGradientBoostingmodelstominimizeprediction variance and improve generalization across diverse symptom patterns. B. PracticalDeployability: Implementingalightweight,web-basedinterfacethatsupportsreal-timesymptominput,diagnosis,andresult interpretation with minimal training, even in low-resource settings. C. Data-DrivenIntervention: Enablingserviceproviderstotrackandanalyzepoisoningtrendsviainteractivedashboards,supportingtimelyhealth interventions and resource allocation. However,systemperformanceisinfluencedby: • Synthetic Data Bias: Limited real-world datasets may introduce generalization challenges when exposed to rare orregion-specific poisoning cases. • User Input Dependency: Accuracy depends on the correctness and completeness of symptom data entered by non-expert users. KeyImprovementsOverTraditionalRuralDiagnosticMethods

References

[1] Carvalho, J. C. S., Pimenta, T. C., Silverio, A. C. P., Carvalho, M. A., & Carvalho, J. P. C.S. (2024). \"A New Data Science Model With Supervised LearningandItsApplication onPesticidePoisoningDiagnosisin RuralWorkers.\" IEEEAccess, vol.12,pp.1–12. [2] Anandhi,G.,&Iyapparaja,M.(2024).\"SystematicApproachestoMachineLearningModelsforPredictingPesticide Toxicity.\" Heliyon, vol. 10, e28752. [3] Mienye, I. D., & Sun, Y. (2023). \"A Survey of Ensemble Learning: Concepts, Algorithms, Applications, and Prospects.\"IEEE Access, vol. 11, pp. 1–25. [4] Tabakova-Komsalova,V.,Nikolov,V.,&Todorov,T.(2023).\"AnExpertSystemfortheDiagnosisofLivestockPoisoning.\" [5] IEEEAccess,vol.11,pp.10123–10131. [6] Al-Kasasbeh,R.T.,etal.(2022).\"FuzzyMathematicalModelsforPredictingandDiagnosingOccupationalDiseasesof Workers in the Agro-industrial Complex in Contact with Pesticides.\" IEEE Access, vol. 10, pp. 150203–150215. [7] Manduca,G.,etal.(2023).\"AutomatedImage-BasedAnalysisUnveilsAcuteEffectsDuetoSub-LethalPesticideDoses Exposure.\" IEEE Access, vol. 11, pp. 76834–76842. [8] B?la?a,D.,&Mih?escu,R.(2023).\"MachineLearningModelsforPredictingAcutePesticidePoisoningBasedonClinical [9] Parameters.\"ComputersinBiologyandMedicine,vol.158,106816. [10] Abdelaziz, A., &Elazab, A. (2022). \"Explainable AI for Toxic Substance Detection and Medical Diagnosis.\"IEEE Journal ofBiomedical and Health Informatics, vol. 26, no. 9, pp. 4325–4336. [11] Patil, S., & Pawar, S. (2021). \"Predictive Modeling of Pesticide Poisoning Using Hybrid Ensemble Techniques in Agricultural Regions.\" International Journal of Environmental Research and Public Health, vol. 18, no. 12, 65523.

Copyright

Copyright © 2025 Sama Prakash Reddy, Goli Karthik, Mandhula Praveen, Bh Bhujanga Reddy. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.