OPTIMISED SUPPORT VECTOR MACHINE (SVM) FOR DETECTION OF ANDROID MALWARE WITH NEIGHBOURHOOD COMPONENT ANALYSIS ALGORITHM

Abstract

ABSTRACT The exponential surge in android malware has continued to increase with new variants of the malware surfacing daily. This is due to the ubiquitous nature of mobile platforms and internet technology which are almost inevitable in today’s digital age. One of the approaches used in detecting malware is the use of machine learning algorithm. SVM is a machine learning (ML) classifier that has demonstrated promising strength to be used as a tool for the detection of android malware. The goal of this study was to build an optimised SVM model for detection of android malware using the neighbourhood component analysis (NCA) algorithm. The research work adopted the neighbourhood component analysis (NCA) algorithm to sieve out irrelevant features which guaranteed excellent model performance. The Bayesian optimization method was used to optimally combine the various SVM hyperparameters and their values to have a super and optimised NCA-BOM-SVM model for the detection of android malware. The results of the proposed model (NCA-SVM) show an accuracy of 97.8%, false alarm rate of 0.021, precision of 97.9%, error rate of 0.02, recall of 97.9%, and f1_score of 97.9%. These results show an enhanced performance of malware classification with higher accuracy and precision, alongside reduced false alarm rate and error rate, thus demonstrating an improvement on existing literature. This implies that the Bayesian optimization of SVM alongside the neighbourhood component analysis algorithm provides a more robust tool for the detection of android malware