An Ensemble Framework for Improving Brain Stroke Prediction Performance

Abstract

Brain stroke detection using data-driven approach has economic benefits. Simple approach using Machine Learning (ML) classification algorithms could provide acceptable accuracy for realizing Clinical Decision Support System (CDSS). From the literature, it is ascertained that making ensemble of multiple brain stroke prediction models could improve prediction performance. This is the hypothesis and motivation for the research carried out and presented in this paper. Another important observation from the literature is that most of the ensemble methods found in the literature for brain stroke prediction are not data-driven approaches. This research gap is filled in this paper by focusing on ensemble of data-driven prediction models. Towards this end, we proposed an ensemble framework based on supervised ML techniques for improving brain stroke prediction performance. The framework is named as Brain Stroke Prediction Ensemble (BSPE). We also proposed an algorithm known as Hybrid Ensemble Learning for Brain Stroke Prediction (HEL-BSP). We also reuse our feature engineering algorithm known as Composite Metric based Feature Selection (CMFS). The ensemble is made up of ML models such as Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), KNeighbours classifier, Gradient Boosting and Stochastic Gradient Descent (SGD). A prototype application is built using Python data science platform to evaluate the proposed framework and the underlying algorithm. The experimental results revealed that the ensemble of the prediction models with majority voting approach could outperform individual prediction models.