Optimization of feature selection in Support Vector Machines (SVM) using recursive feature elimination (RFE) and particle swarm optimization (PSO) for heart disease detection

• Main Authors: Bayuaji, Luhur, Kusnadi, Kusnadi, Amzah, Mochamad Yamin, Pebrianti, Dwi
• Format: Proceeding Paper
• Language: English; English
• Published: IEEE 2024
• Subjects: T Technology (General)
• Online Access: http://irep.iium.edu.my/115360/1/115360_Optimization%20of%20feature%20selection.pdf; http://irep.iium.edu.my/115360/2/115360_Optimization%20of%20feature%20selection_SCOPUS.pdf; http://irep.iium.edu.my/115360/; https://ieeexplore.ieee.org/document/10652561
• Institution: Universiti Islam Antarabangsa Malaysia

Heart disease is one of the main health problems throughout the world. Early and accurate detection of heart disease is very important to reduce the death rate caused by this condition. One effective approach to detect heart disease is to use Support Vector Machine (SVM) as a machine learning algorithm. However, when using SVM, selecting the right features is very important to improve prediction performance. Inappropriate feature selection can lead to excess dimensions, high computing time, as well as the possibility of adding irrelevant information. Apart from that, the choice of C and gamma parameters can affect SVM performance. The RFE method is used to select the most informative and relevant subset of features from a given feature set. RFE combines a dimensionality reduction process with a machine learning process where the least important features are iteratively removed until the best feature subset is obtained, while the PSO method is used to optimize the C and gamma parameters of SVM. The results of the SVM-RFE PSO model trial showed increased accuracy in classifying heart disease. The accuracy value increased from 86.41% to 89.13%, indicating that this approach compares favorably with conventional SVM models or even SVM with RFE alone. These results illustrate that using the combination of RFE and PSO together is effective in improving the ability of SVM in classifying heart disease.