Design And Performance Optimization Of Android And Windows Malware Detection Systems Using Machine Learning and Evolutionary Computing Techniques

Abstract

ABSTRACT newlineMalware poses a significant threat to information, networks, and systems, compromising confidentiality, integrity, and availability. Mobile platforms, notably Android, and Windows for desktop computers are particularly vulnerable to malware attacks. This research aims to improve malware detection in these systems through the use of innovative methods such as statistical feature selection, fuzzy meta-heuristic optimization, and machine learning classifiers. newlineThe study focuses on identifying the most influential Android permission attributes and improving performance using statistical methods and machine learning classifiers. Additionally, the research combines fuzzy logic with swarm optimization for more efficient Android malware detection and classification. The study also explores ways to optimize API data using auto-encoders and Artificial Neuronal Classifier combined with optimization algorithms for Android malware detection. newlineMoreover, the research develops swarm optimization and machine learning wrapper models for dynamic Android malware detection based on API calls. Furthermore, this research introduces a new dataset for Windows PE malware detection called SOMLAP. The results show that the innovative methods used in this study can significantly improve malware detection accuracy while reducing feature datasets. newlineThis research also focuses on feature selection and filter-based classifier for static Android malware detection achieving an accuracy of 93.5%. The study also developed a two-tier fuzzy meta-heuristic optimization for dynamic Android malware detection achieving an accuracy of 98.53% with a 95% reduction in the feature dataset. Furthermore, a swarm optimization and machine learning wrapper model for dynamic Android malware detection achieved an accuracy of 98.87%, reducing the feature dataset to only seven features out of 100. Lastly, the study achieved the best result in Windows PE malware detection with an accuracy of 99.37% after reducing the feature dataset to only twelve out of 108 featur