A Classification and Regression Tree Based Predictive System for Preventing Stroke

Abstract

Everybody wants to be healthy, but sometimes illnesses and disorders prevent this from happening. Most deadly diseases can be treated or their harmful effects can be reduced if detected at primary phase(s). Stroke is a silent killer and a very deadly disease which kills 17.9 million people yearly according to World Health Organization. It typically happens as a result of an obstruction or hemorrhage in the blood vessels that provide blood to the human brain. This impairs or reduces the blood flow to the brain, which prevents it from receiving enough oxygen or nutrients, ultimately leading to the death of brain cells. Over the years, the applications of data-mining techniques have been remarkable in the field of medicine. This has led to progressive improvement in the areas of diagnoses, predictions and intensely understanding of healthcare data. Stroke dataset gotten from kaggle.com were used in this work. Classification and Regression Tree (CART) technique was employed to develop a predictive model that could predict people at risk of having stroke. The predictive model was trained on 43,000 records, and it was tested on 19,000 records. Detection rates of 100% and 98.42%, respectively were achieved by the model's prediction accuracies when tested on training and testing sets. The model was implemented using Python programming. It is hopeful this work will be of positive contribution in the health sector.