Impact of SMOTE and Feature Selection on SVM and KNN Performance in Smoker Classification

Abstract

This research classifies the smoker status dataset using two algorithms, i.e., Support Vector Machine (SVM) and K-Nearest Neighbors (KNN). Researchers used several kernels in the SVM model: linear, RBF, and Polynomial. Researchers implemented K values of 3 and 110 in the KNN model. This study uses two data types: the original data of 55692 rows and the data balanced by applying SMOTE of 70474. The evaluation results show that the linear kernel SVM method with the SMOTE process and forward selection method produces the highest accuracy of 76%. In comparison, the accuracy of the KNN method with the SMOTE process and using all features get the highest accuracy of 84%. This research shows that the application of SMOTE can improve model performance. In addition, the application of feature selection also affects the accuracy results except for the KNN model on SMOTE data using all features.