Investigating Key Predictors and Validating a Predictive Intelligence Model for Non-Communicable Disease Risk: Case Study Kitui County, Kenya

Non-communicable diseases (NCDs) account for more than 70% of the global mortality, with an increasing burden in low-resource environments such as the Kitui County in Kenya. This paper examined the predictors of NCD risk that are considered important and tested using a predictive machine learning model with 68,601 anonymized patient records. The study identified systolic blood pressure (𝑟 =0.72), body mass index (𝑟 = 0.67) and fasting blood sugar (𝑟 = 0.58) are most influential predictors. A hybrid ensemble model consists of Random Forest and XGBoost achieved an accuracy of 93% with a precision of 0.81, recall of 0.82 and an F1-score of 0.81, validated through 5-fold cross-validation. SHAP analysis enhanced interpretability by providing clinically relevant, patient level insights consistent with medical reasoning. The model demonstrated a high predictive rate with a reduced false negative rate, hence assisting in early detection and intervention. These results demonstrate the potential of interpretable machine learning models in scalable and effective NCD screening in resource-constrained healthcare.