Optimizing Malware Detection and Classification in Real-Time Using Hybrid Deep Learning Approaches

Abstract

Malware detection and classification are critical for ensuring system security in real-time applications. Conventional approaches may not be optimized to combine precise results with low time consumption and become a problem when it comes to processing large volumes of different malware samples in a real-time setting. The general framework for this paper is to introduce a new detection and classification method that uses deep learning (DL) models to detect and classify malware. We developed and tested two models: the static convolutional neural network-long short-term memory (CNN-LSTM) model and the dynamic CNN 1D-LSTM model in this work. The models achieved an accurate rate of 99%. Static-CNN-LSTM was able to classify the malware based on static analysis. At the same time, the proposed dynamic (1D-CNN-LSTM) model got the best results, with a 100% success rate, by gathering behavioral data. This means that it can accurately classify even new and complicated dynamic malicious program variants. Therefore, this study's results show that using a hybrid approach raises the rate of detection while also meeting the real-time processing needs of systems with a lot at stake that need to perform well. Our approach represents a substantial improvement in malware detection, delivering a more efficient and versatile response to contemporary cyber threats.