Jaccard-ABC e Cor-ABC para seleção de características em datasets de malware Android

Abstract

This dissertation aims to improve feature selection in datasets focused on Android malware detection by proposing two new neighborhood exploration methods based on similarity measures for the Artificial Bee Colony (ABC) algorithm. The proposed meth- ods, Jaccard-ABC and Cor-ABC, demonstrated effectiveness in reducing the number of features across various scenarios, often surpassing 30% reduction, with Jaccard-ABC outperforming the traditional ABC in all scenarios. Furthermore, the methods main- tained competitive performance in terms of accuracy, precision, recall, and F1, with variations generally less than 1% compared to the traditional ABC. The Cor-ABC method, in particular, exhibited greater stability by reducing internal randomness in the selection process. In conclusion, the proposed methods provide valuable contributions to efficient feature selection, especially in the context of malware detection on Android devices.