Performance Evaluation of Machine Learning Techniques for Early Prediction of Brain Strokes

Abstract

Brain Strokes are the prime cause of fatality in the world. Bangladesh probably has the highest rates of brain strokes among all South Asian countries and yet is the least studied. Predicting stroke effect from a set of predictive attributes may classify high-risk patients and guide cure approaches, leading to reduce relative incidence. In this work, we propose an intelligent system that can make an effectively prediction of a possible brain strokes using only eight (8) features. We also apply six (06) well known supervised machine learning algorithms on first Bangladeshi datasets collected from five different hospitals of Bangladesh to analyze the prediction accuracy. The overall process can be categorized into four phases. Phase 1: we have provided a comprehensive literature review where we summarize various related machine learning algorithms. Phase 2: we have collected brain stokes patients’ data from five different hospitals of Bangladesh to create a dataset. Phase 3: we have selected the important features by using feature importance score. Finally, feed the data to appropriate machine learning algorithms to determine if the predictive model is accurate. It is observed that using our collected dataset for 8 features the classification accuracy of Bagging is almost 96% and it performs better than other classification algorithms such as Logistic Regression (94.82%), k-Nearest Neighbor (73.27%), support vector machines (93.96%), Naive Bayes (93.97%) and Decision Tree (89.66%). Whereas using the dataset with all 23 attributes the classification accuracy of Bagging is 93.43% and it also performs better than the other classification algorithms, such as Logistic Regression (92.24%), k-Nearest Neighbor (69.83%), support vector machines (90.52%), Naive Bayes (92.24%) and Decision Tree (87.07%).