Dynamic Malware Detection Using Effective Machine Learning Models with Feature Selection Techniques

Dynamic Malware is a type of virus that is self-modifying, which makes it difficult to analyze in the course of its operation. It occasionally changes its behavior based on the existing environment and the context of execution. The goal of this study was to identify and detect dynamic malware in Android devices using effective machine-learning models with feature selection techniques. With new malicious software emerging daily, relying solely on manual heuristic analysis has become ineffective. To address this limitation, the study used dynamic detection methods to detect the events of interest using machine learning models. Some of these measures entailed duplication of an environment in which the behavior of malware could be replicated and then come up with reports. The reports were then transformed into sparse vector models so that other machine-learning techniques could then be applied to them. In this research study seven different models, namely, KNN, DT, RF, AdaBoost, SGD, Extra Trees, and Gaussian NB, were used to train an effective malware detection model to predict the dynamic malware in its early stages. The study showed that Random Forest, Stochastic Gradient Descent, Extra Tree, and Gaussian Naive Bayes classifiers achieved the highest accuracy compared to other models. This study endorses the application of machine learning-based automated behavior analysis for malware detection, about the complexities involved in the dynamic behavioral analysis of malicious software.