A Machine Learning-Based Health Issues Detection System in Crude Oil Exploitation Host Communities in Ondo State

Abstract

Crude oil exploitation in Ondo State, Nigeria, causes severe environmental pollution and adverse health outcomes like respiratory disorders and water-borne illnesses. Current surveillance systems remain reactive and inefficient at detecting early warning signs. To address this gap, this study developed a machine learning-based health issues detection system (HIDS) to enhance early detection in vulnerable populations. Due to data collection constraints, synthetic dataset of 5000 records was generated based on epidemiological patterns and environmental indicators including air pollution, water contamination, and proximity to gas flare sites. The Random Forest algorithm was selected for its robustness with non-linear relationships. Evaluated using an 80:20 split, the model achieved strong performance: accuracy 82.34%, precision 83.12%, recall 81.45%, specificity 83.56%, F1-score 82.23%, and ROC-AUC 85.23%. Feature imporatnce analysis identified environmental factors as dominant predictors. Air pollution index (0.287), water contamination index (0.245), and distance to gas flare sites (0.198) collectively accounted for 73% of predictive power. Additionally, a three-tier risk stratification framework was established where low (0.00-0.40), moderate (0.41-0.70), and high (0.71-1.00) to facilitate proactive healthcare interventions and efficient resource allocation. This paper confirms the feasibility of applying machine learning for public health surveillance in resource constrained environments. It provides policymakers and healthcare providers with vital tool for mitigating health risk in crude oil host communities through data-driven early warning systems rather than reactive approaches, ultimately improving health outcomes in environmentally vulnerable populations affected by extractive industry activities.