基于集成机器学习算法的Android恶意软件创新预测方法研究

Research on Android Malicious Software Innovation Prediction Method Based on Integrated Machine Learning Algorithms

恶意软件旨在破坏、禁用或控制计算机系统.Android恶意软件专门针对Android操作系统,以泄露机密信息和破坏系统为目的.文献显示相关领域已进行了多次尝试来检测Android恶意软件.然而,这些工作无法自动检测恶意软件,而且大多数都是基于签名的,无法检测恶意软件的新变种.本研究中,探索了不同的算法,以获得恶意软件预测的最佳算法,并获得有助于本研究有效预测恶意软件的最佳特征集.从本研究的分析中,已经看到,在预测恶意软件方面,集成方法比传统的机器学习算法要好.本研究使用LGBM创新算法将特征数量从215个减少到100个,精准率达到99.50%.此外,本研究使用只有55个特征的随机森林获得了99.17%的精准度。

Malicious software aims to destroy,disable,or control computer systems.Android malware is specifically targeted at the Android operating system,with the aim of leaking confidential information and damaging the system.The literature shows that multiple attempts have been made in the relevant field to detect Android malware.However,these tasks cannot automatically detect malware,and most of them are signature-based,making it impossible to detect new variants of malware.In this study,different algorithms were explored to obtain the best algorithm for predicting malware and to obtain the best feature set that can help effectively predict malware.The analysis of this study shows that ensemble methods are better than traditional machine learning algorithms in predicting malware.The LGBM innovative algorithm is used in this study to reduce the number of features from 215 to 100,with an accuracy rate of 99.5%.In addition,an accuracy of 99.17%is achieved using a random forest with only 55 features.