Android malware prediction based deep learning approaches: dimensionality reduction and data transformation. رسالة ماجستير

Abstract

Android is free, open-source and the most popular mobile operating system. Android's worldwide market share was 72.22% in the fourth quarter of 2020, and although it dropped to 71.8% by the end of 2022, it is still well in front. Recent Android malware defenses, which detects dangerous data of malware based machine learning, have become a significant issue in information security research due to their importance in keeping devices secure. Traditional machine learning approaches are limited in their ability to learn complicated representations in high-dimensional domains. Furthermore, the success of machine learning models relies heavily on training data, and as Android apps evolve and software engineering advances, these trained models are likely to become outdated. This research develops a hybrid Android app classification model into benign and harmful based Convolutional Neural Networks. The model combines dimensionality reduction and data conversion into images using Image Generator for Tabular Data algorithm. To evaluate the model, the CICMalDroid 2020 data set was used. The proposed model achieved a high accuracy of 94.38% in the binary classification (benign and harmful) and 95.83% in the multiple classification.