An Optimized SVM with Feature Selection Using Swarm Intelligence Technique

Support Vector Machine (SVM) is one of the most older machine learning algorithms, but it is widely used so far to solve both classification and regression problems with a good practical results. As any other learning algorithm, the SVM performance depends on the relevance of the input features. However, this performance can be further improved by adjusting some of SVM parameters. Under this context, this paper introduces the SCSO+SVM algorithm to optimize the SVM parameters, and in the same time, to perform an effective feature selection for classification tasks by using a recent swarm intelligence technique called Sand Cat Swarm Optimizer. Experiments were conducted first on six benchmarking datasets using the proposed SCSO+SVM and four SVM-based algorithms, and then a disease Covid-19 dataset is used to verify the applicabilty of the algorithm on more recent and real-world problems. The results show a better performance of the SCSO+SVM algorithm in regards to selecting the relevant features and optimizing the SVM accuracy.