GA-based feature selection for improved malware detection

Malware or malicious software can take any form, infect any machine and cause any extent of damage to the organisational data and processes. Malware detection is so the most important task of any security team, and I believe the faster we detect it, the more we can reduce the damage drastically. Machine learning improved malware detection and classification problems with better accuracy. These algorithms can be trained effectively with a large pool of malware data. But as the data increases, the feature overfilling becomes an adversary to these classifiers. This research checks whether feature reduction can improve the classification accuracy of the algorithms. This model uses a Genetic Algorithm for the feature selection as it overpowers another algorithm like PSO, ACO, etc., with its simplicity and effectiveness. A suit of 11 algorithms is evaluated with GA-optimized features and the implementation is tested on a sample of Microsoft Malware dataset. The results show an increased accuracy for classification and enforce the fact that feature reduction can improve classification accuracy. The proposed method is novel in terms of the performance evaluation of the algorithms after feature selection. The result indicate the algorithms which are sensitive to feature selection and which are not, indicating the suitability of combining feature selection with these algorithms.