Machine Learning Approaches for Threat Detection in Android Mobile Networks

Abstract

The development of mobile technologies and the high number of applications based on Android have considerably increased the number of targets of attack by cybercriminals. Android, which is the most popular mobile platform, is becoming the target of more and more malware and adware that aim to steal the privacy, financial data, and integrity of the system of its users. The paper suggests a complex Android malware detection system based on an Artificial Neural Network (ANN) that can be used to improve the classification of malicious and innocuous applications. The study is based on a sample of 4,464 Android instances and includes necessary preprocessing steps, including handling missing values and outliers, min-max normalization, and feature selection using PCA. Random under-sampling was used to overcome the imbalance in classes prior to training of ANN model. The offered ANN has better results, with the accuracy of 99.38%, precision of 99.24%, recall of 99.42%, and F1-score of 99.61%. Accuracy loss curves, confusion matrix, and ROC analysis evaluation have verified the high generalization of the model, learning stability, and good discriminative properties. Comparative findings also indicate that the ANN performs more successfully than traditional models which include, LSTM (93.9%), SVM (94%) and CNN (97.8%). All in all, the research offers a viable deep learning-driven source of strong Android threat detection, which enhances the safety and well-being of the mobile context