Development of an Intelligent System for Brain Stroke Prediction using Ensemble Feature Selection and Machine Learning Technique

Brain stroke is a serious condition in human body, in which cerebral artery ruptures causing damage to the region and preventing oxygen and other substances from reaching the brain. The risk of brain stroke can be reduced dramatically by early detection through the analysis of various stroke warning indicators in the patient’s body. In such cases, predictive analytic of brain stroke through an ensemble feature selection and machine learning using patients’ various clinical symptoms can be an excellent solution. The goal of this research is to develop a predictive intelligent system that would estimate the possibility of occurring brain stroke using patient symptom information applying machine learning techniques. To attain the objective, the relevant patient dataset has been collected, analyzed and visualized. The dataset has undergone rigorous preprocessing using multiple data exploration techniques. Five different feature selection method are followed by an ensemble methodology with the maximum voting approach to extract the most important feature and generate a dataset with reduced attributes. Following that, ten types of machine learning classifiers are trained, examined and put to the test to predict the likelihood of brain stroke. Each of the classification models are evaluated using multiple performance parameters. The algorithm that handled this challenge the best is Logistic regression model, which has provided an accuracy of almost 82.7% employing the dataset with reduced clinical features for stroke prediction.