Use of Machine Learning Algorithms for Android App Malware Detection
Shaurya Rawat, Rushang Phira, Prachi Natu · 2021 5th International Conference on Electrical, Electronics, Communication, Computer Technologies and Optimization Techniques (ICEECCOT) · 2021
Malware attacks in mobile phones have become more frequent with time. One of the most common ways to infect a phone system with malware is through apps that look benign but are actually malicious. Machine Learning has been employed in the past to detect malware in phones. This has been achieved through supervised learning algorithms and Deep Learning. This paper analyses the performance of various classification algorithms to detect malware in android apps based on permissions requested from the user. The intention was to find as accurate a model as possible to make predictions based on the type of android package. Through implementation of multiple models, XGBoost classifier was found to yield the highest accuracy with 80.03%, followed by Cat Boost Classifier and Random Forest Classifier with 79.58% and 78.67% respectively. Even in most other metrics, XGBoost classifier gave the best results. The data was thoroughly cleaned and the relevant preprocessing steps were implemented, following which classification was performed along with detailed hyperparameter tuning. Various metrics were used to compare the results between the different algorithms, which were accuracy, precision, recall, F1 score and logistic loss function. The findings could help make people aware about the kind of permissions required of malicious applications.