Development of an Explainable Machine Learning Model For Hypertention Risk Prediction Using Basic Clinical Parameters

Abstract

Hypertension is one of the major contributors for gravid women and unborn child’ death especially in the developing countries where access to innovative diagnostic technologies is not common. The study used explainable machine learning algorithm to predict hypertension risk using basic clinical attributes. Pregnant women record in the Mother and Child Hospital in Akure was collected and preprocessed to address missing values, regulate the features, and enhance overall data quality. The data partitioning at 80% for training and 20% for testing was carried out to confirm a balanced representation of the target variable. Predictive models were developed and evaluated to measure their forecasting performance. Random Forest model achieved 90% accuracy and was able to classify both “HIGH and “LOW” risk level. In addition, SHapley Additive exPlanations (SHAP) analysis was adopted to select the most influential predictors in the model and the result identify systolic Bp, Temperature, Diastolic and weight as most factors contributing to hypertension risk. In conclusion, the developed model can serve as an intelligent clinical decision support tool to assist healthcare professionals in the early detection of hypertension risk using readily available clinical data.