Ensemble Machine Learning for Stroke Risk Prognosis: Leveraging Clinical Data in a Nigerian Healthcare Setting

Stroke, a nervous system disorder usually depicted by blood vessel blockage, is one of the acute global diseases, the third leading cause of death and disability worldwide, that becomes prevalent with age, has resulted in a high economic burden, long-term disability, and high mortality rate, affecting about 15 million people yearly. Over 100 million people have experienced a stroke, with 12 million people projected to have their first stroke in 2026, and 6.5 million deaths projected. It was also recorded that 1 in 4 people over 25 years of age will experience a stroke in their lifetime. Hence, there is a need for prompt diagnosis of susceptible and endangered individuals for timely intervention and appropriate treatment. This study presents an ensemble stroke prediction machine learning model that utilizes clinical datasets obtained from the University of Medical Sciences Teaching Hospital in Akure, Nigeria. Scaling, Label encoding, and Synthetic Minority Over-sampling Technique (SMOTE) were all utilized in the dataset preprocessing stage for proper data augmentation. The study harnesses the joint advantages of K-Nearest Neighbours (KNN) and Support Vector Machines (SVM) combined using a soft voting ensemble technique. The system attained 98% accuracy, 98% precision, 95% recall, and an AUC of 1.00. The system’s result is evidence that ensemble learning can be easily deployed in early stroke diagnosis, especially in clinical settings with limited resources.