Analyzing Texture-Based Feature Extraction and Classification Techniques for Malware Detection

Malware detection is a vital part of cybersecurity, yet challenges such as low accuracy, high computation, and heavy resource usage persist. This work introduces a detection framework that integrates multiple texture-based feature extraction methods—SFTA, LBP, Haralick, Gabor, and Tamura—for analyzing malware images. Experiments were conducted on the MalImg and MaleVis datasets using various classifiers, including RF, KNN, GDA, SVM, LR, ELM,andan ensemble Voting Classifier combining Boosted Trees, Bagging with RF, and LR. The results show that GDA achieved 91.3% accuracy on MaleVis, while ELM reached 96.7% on MalImg. These findings emphasize the need for effective feature selection and classifier choice to improve malware detection performance.