Optimization of Stroke Prediction Model using a Hybrid of Heterogeneous Ensemble Machine Learning Techniques

Abstract

Stroke is a chronic disease caused by poor blood flow into some brain cells, oxygen and nutrient deficiency, causing the cells to die. The cure and prediction of the occurrence of stroke have been a problem. Machine learning techniques have of late proven to be efficacious in disease predictions. This paper utilizes a hybrid of three heterogeneous ensemble machine learning models: Logistic Regression (LR), Support Vector Machine (SVM) and Random Forest (RF) yielding SVM-LR-RF for improved stroke using healthcare dataset obtained from Kaggle repository. The models were trained and evaluated in Python programming environment using confusion matrix and soft voting. The prediction results of the base learners served as inputs to the hybrid SVM-LR-RF model. The results showed that the hybrid model performs better with a prediction accuracy of 95%, compared to the individual base learning models. It is thus recommended for use for more effective stroke prediction.