Desarrollo de clasificador multiclase de malware ofuscado mediante redes neuronales y modelos estadísticos

Abstract

This monograph presents the development of a multiclass classification model for obfuscated malware using neural networks and statistical models. Given the growing challenge that malware poses to cybersecurity, especially due to the sophistication of techniques such as obfuscation, advanced methods for its detection and classification are explored. The study focuses on identifying and analyzing essential attributes for malware classification based on data extracted from computational networks. Several ensemble models, including random forests, Adaboost, XGBoost, and convolutional and feed-forward neural networks, are implemented and evaluated. The performance of these models is considered when applying preprocessing techniques, such as feature selection and data augmentation through the addition of white noise. The monograph concludes with a proposed model that combines multiple estimators to improve the accuracy of detecting different types of malware, such as trojans, ransomware, and spyware. This approach offers a significant advancement in the protection of electronic systems, contributing to cybersecurity in both industrial and governmental environments.