Design of an efficient Malware Prediction Model using Auto Encoded & Attention-based Recurrent Graph Relationship Analysis
Mahesh T. Dhande, Sanjaykumar Tiwari, Nikhil J. Rathod · International Research Journal of Multidisciplinary Technovation · 2025
The threat of modern malware in the world of cyber security has grown and how the need for proper detection and analysis techniques has grown with it. All these conventional approaches are insufficient methods if used to detect new or emerging strains of malware. For this need, the present research develops a novel Malware Prediction Model using Auto Encoders and Attention Mechanisms to advance Malware Pattern Analysis. This new approach goes beyond the conventional wisdom because it decodes complex patterns of malware into identifiable Malware Classes utilizing the unique Recurrent Graph Relationship Analysis. Recurrent Networks perform the complex task of Feature Analysis and simultaneously. Classical approaches mainly conceive pattern matching where signatures are taken and used to look in the system hence cannot detect polymorphic or metamorphic types of viruses. Additionally, these systems have high levels of false positives and poor ability to learn from new types of threats. On the other hand, the coupling of Auto Encoders with Attention Mechanisms in the model under consideration allows the model to gain better insights of malware behavior. Such an integration not only improves the identification of multiform patterns but also changes the approach to growing threats more effectively. The use of this model was benchmarked against two databases: The Malware Memory Analysis and The Kharon Malware Database Samples. Strikingly, the proposed model provided 8.3% more precision, 8.5% more accuracy, 5.9% higher recall, 6.5% better AUC, higher specificity by 9.4%, while slight reduction in delay by 2.9% to other methods.