Análise e aplicação de modelos de inteligência artificial para detecção de malware em dispositivos móveis Android

Abstract

The study is driven by the increasing number of cyber attacks on Android systems, given their widespread use and the sensitive data stored in them. The work aims to evaluate the efficiency of artificial intelligence (AI) models in detecting malware from Android devices using the CIC- MalDroid 2020 dataset. The models that were applied include AdaBoost, Gradient Boosting, and KNN. Others are Logistic Regression, MLPClassifier, Random Forest, SVM, and XGBoost. The acquired dataset provided the static and dynamic features of applications. Thus, machine learning techniques could be used to detect behaviors associated with malicious software. In summary, XGBoost, Gradient Boosting, and Random Forest models produced the best results for various types of malware such as Adware, Banking Malware, SMS Malware, and Riskware. This is quite enlightening towards getting better solutions in mobile security. Their respective F1- scores were 0.945577, 0.937279, and 0.937227. Conversely, the SVM, Logistic Regression, and AdaBoost models exhibited the lowest performance, providing F1-scores of 0.859231, 0.838185, and 0.796414, respectively.