Malware detection using machine learning and feature selection methods

This thesis aims to develop effective prediction models for malware detection using various machine learning methods and feature selection methods. Specifically, the study utilizes Light Gradient Boosting Machine (LightGBM), Decision Tree (DT), K-Nearest Neighbors (KNN), Multi-Layer Perceptron (MLP), Support Vector Machines (SVM), and Naive Bayes (NB) classifiers, each combined with Relief-F and Minimum Redundancy Maximum Relevance (mRMR) feature selectors. The analysis is conducted on 5136 samples, comprising 2683 malicious and 2501 benign files, derived from PE file header information of the Windows operating system. Evaluation of model performance is based on accuracy, F1-Score, recall, and precision criteria. Furthermore, to enhance the generalization ability of each model, 10-fold cross-validation is employed. The complexity matrices of the best performing model for each method are also provided. The results indicate that the LightGBM algorithm demonstrates promising outcomes in malware detection, both with default and optimized parameters. Moreover, the Relief-F feature selector exhibits superior performance compared to mRMR. These findings underscore the potential of LightGBM as an effective approach in malware detection and suggest that techniques such as Relief-F has significant potential for feature selection in this particular area.