SNEL-DFF: Android malware detection using Siamese networks with ensemble learning

Abstract

This paper proposes a new model simply known as Siamese Networks of Optimal Ensemble Learning with Deep Forest Feature (SNEL-DFF). The proposed model has the Deep Forest Feature extraction feature because of the complexity that is present in the data and to enhance the proficiency of the detection system. The feature vectors used in this study includes 215 attributes in android applications which are derived from samples sourced from Drebin dataset. Some of the performance evaluation results have been highlighted revealing that the proposed model yielded an accuracy of 0.99. The precision of 0.98 shows its ability to avoid miss-identification of negatives and the recall of 0.99 proves the effectiveness of using it for detection of the real malware samples. The F1 score is 0.99 and ROC-AUC value of 0.99 indicating the model has achieved 99% accuracy which points to the fact that it is balanced and provides a superior performance. These findings vindicate the hypothesis that SNEL-DFF has strong predictive accuracy as compared to the conventional machine learning algorithms. The proposed technique utilizes Siamese networks, deep forest feature enhancement, and ensemble learning, which makes it perform better than its competitors in terms of various evaluation criteria.