Implementation of particle swarm optimation-support vector machine with SMOTE for stroke classification

Abstract

Various technologies are applied to improve the quality of services in the health sector. Several studies applied classification techniques and found the importance of using artificial intelligence (AI) methodologies in healthcare systems, particularly disease identification. This research aims to determine the results of stroke classification using the Support Vector Machine method and the differences in accuracy of stroke classification using the Support Vector Machine and Particle Swarm Optimization-based Support Vector Machine methods. Before classifying the stroke dataset, preprocessing is carried out first, such as dealing with unbalanced data using the SMOTE technique so that the data is ready to be entered into the classification model. The stroke dataset has nine features or attributes, 5,510 data before and 9,772 data after upsampling. This research obtained a model for a simple prediction system for stroke classification and obtained accuracy values of 88% using the SVM method with the SMOTE technique and 95% for testing using the PSO-SVM method with the SMOTE technique.