LINEAR CLASSIFICATION MODELS AND TREE CLASSIFIER APPROACH IN THE DETECTION OF ANDROID MALWARE WITH HYPER-PARAMETER TUNING

ABSTRACT

The increasing sophistication and amount of malicious activities on users by mobile malware has resulted from the continued global adoption of smartphones, subverting what has become the largest range of targets with the most lucrative rewards. Malware detection is critical because of the types of harm that can be caused, such as unauthorized data extraction and dissemination, user spying, and so on. This research explores hyper-parameter tuning of Linear and tree-based Scikit-Learn machine learning algorithms using the random search technique as a means to optimize and increase the accuracy of malware detection models. A wide range of hyper-parameters were investigated in order to find the optimal configurations for malware detection. The random search technique was tested with the Random Forest and Decision Tree classifiers, as well as the Linear Models; Logistic Regression and Support Vector Machine (SVM). The performance metrics used are Correctly Classified (accuracy), True Positive Rate (TPR), False Positive Rate (FPR), Precision, and Area under the ROC curve (AUC) score. SVM model gave an increased accuracy of 98.69% with optimal configuration of C=100, gamma = 0.01 and kernel = ‘rbf’.

Keywords: Android: Malware, Hyper-Parameter Tuning, Classifier.