Stroke Prediction Using Machine Learning in a Distributed Environment

Abstract

As with our changing lifestyles, certain biological dimensions of human lives are changing, making people more vulnerable towards stroke problem. Stroke is a medical condition in which parts of the brain do not get blood supply and a person attains stroke condition which can be fatal at times. As these stroke cases are increasing at an alarming rate, there is a need to analyze about factors affecting the growth rate of these cases. There is a need to design an approach to predict whether a person will be affected by stroke or not. This paper analyse different machine learning algorithms for better prediction of stroke problem. The algorithms used for analysis include Naive Bayes, Logistic Regression, Decision Tree, Random Forest and Gradient Boosting. We use dataset, which consists of 11 features such as age, gender, BMI (body mass index), etc. The analysis of these features is done using univariate and multivariate plots to observe the correlation between these different features. The analysis also shows how some features such as age, gender, smoking status are important factors and some feature like residence are of less importance. The proposed work is implemented using Apache Spark, which is a distributed general-purpose cluster-computing framework. The Receiver Operating Curve (ROC) of each algorithm is compared and it shows that the Gradient Boosting algorithm gives the best results with the ROC area score of 0.90. After fine-tuning, certain parameters in Gradient Boosting algorithm like optimization of the learning rate, depth of the tree, the number of trees and minimum sample split. The obtained ROC area score is 0.94. Other performance parameters such as Accuracy, Precision, Recall and F1 score values before fine-tuning are 0.867, 0.8673, 0.866 and 0.8659 respectively and after fine-tuning the values are 0.9449, 0.9453, 0.9449 and 0.9448 respectively.