Enhanced Malware Detection using Convolutional Neural Network with Robust Ensemble Algorithm-"LIGHTGBM"
Satya Sandeep Kanumalli, Neha Dalavayi, Varsha Veernapu, Poojitha Kota, Ratna Priya Katari · 2024
This In the past few years use of the Internet to transfer data has increased rapidly. It is becoming the target of malware that exploits the vulnerabilities of the security system. The rapid evolution of malware threats demands innovative solutions for their detection and classification. In response to this challenge, we present a comprehensive approach that harnesses the power of deep learning and ensemble techniques. Our methodology begins with the utilization of a Convolutional Neural Network (CNN) for feature extraction from malware images sourced from the Malimg dataset, which encompasses 25 distinct classes of malware. Through meticulous data preprocessing, including image resizing and normalization, we ensure a standardized and efficient pipeline. The extracted features serve as rich representations of the malware images, capturing intricate patterns learned by the CNN. We then transition to the ensemble phase, incorporating renowned algorithms such as XGBoost, Random Forest, and LightGBM. Hyperparameter tuning optimizes the ensemble models for classification accuracy and computational speed. Rigorous model evaluation, employing metrics such as accuracy, precision, recall, F1-score, and AUC-ROC, reveals insights into the capabilities of our approach. In a landscape where accuracy and efficiency are paramount, our work sets out to provide a holistic solution for the efficient and accurate classification of a diverse range of malware. The Random Forest model would achieve an accuracy of 95%, XGBoost would reach 97%, and LightGBM would attain a remarkable 98%. The computational speed of Random Forest is 58.5%, XGBoost is 29.3% and lightGBM is 12.2%.