Thermamal: Multi-Agent Behavioral Heatmapping for Efficient Malware Detection Using Convolutional Neural Networks

Malware detection remains a critical challenge due to the trade-off between computational efficiency and detection accuracy, particularly in dynamic analysis. Traditional methods often suffer from sequential execution bottlenecks and lack interpretability. This paper introduces ThermaMal, a novel malware detection framework that integrates multi-agent behavioral monitoring with thermal footprint visualization and convolutional neural networks (CNNs). ThermaMal employs specialized agents to monitor memory, disk, registry, network, processes, API calls, hardware, and user interactions in parallel, generating composite heatmaps that encode malware behavior patterns over time. These heatmaps serve as input to a CNN-based classifier, enabling high-accuracy detection while preserving explainability. Experimental evaluations on 5800 malware samples and 5800 benign programs demonstrate that ThermaMal achieves 98 % detection accuracy with a 3 x speedup compared to sequential dynamic analysis. The heatmaps also enhance interpretability by visually distinguishing malicious behavior patterns. By bridging the gap between agent-based monitoring and visual malware analysis, ThermaMal offers a scalable and efficient solution for real-time malware detection.