A HYBRID DEEP LEARNING FRAMEWORK FOR MALWARE DETECTION AND CLASSIFICATION USING STATIC AND DYNAMIC ANALYSIS

Abstract

The rapid proliferation of malware poses a critical threat to modern computing infrastructures, demanding robust and adaptive detection mechanisms. Traditional signature-based antivirus solutions are increasingly inadequate against zero-day exploits and polymorphic malware variants. In this paper, we propose a novel hybrid deep learning framework that integrates static and dynamic malware analysis features with a Convolutional Neural Network–Long Short-Term Memory (CNN-LSTM) architecture. Static features are extracted from Portable Executable (PE) headers and n-gram byte sequences, while dynamic features are derived from API call traces captured in a controlled sandbox environment. Feature fusion is performed through a multi-modal attention mechanism that adaptively weights the contribution of each analysis modality. Evaluation on the MalImg and EMBER benchmark datasets demonstrates that our framework achieves 99.23% accuracy, 99.10% precision, 99.36% recall, and an F1-score of 0.992, outperforming state-of-the-art baselines including Random Forest, Support Vector Machines, standalone CNNs, and LSTMs. These results affirm the efficacy of combining static and dynamic features within a unified deep learning pipeline for real-world malware detection.