Optimization of Deep Learning-Based Ransomeware Detection Using Numeric Feature Analysis with Convolution Neural Network Models

Abstract

The research introduces ResMalNet, a convolutional neural network architecture designed for malware detection. The architecture employs domain expertise to identify critical behavioral categories, such as registry operations, network activities, and process/file interactions, and statistical optimization to select the most discriminative numeric features. ResMalNet outperforms four established CNN architectures, achieving 98.91\% accuracy and 98.92\% precision while maintaining balanced recall and F1-scores of 98.91\%. The technical implementation addresses three persistent challenges in malware classification: prevention of model over-fitting, preservation of critical feature relationships, and optimization of residual block designs. Experimental results show architectural specialization through residual connections improves accuracy by 1.82\% over conventional CNN designs, domain-informed feature selection reduces false positive rates by 42\%, and exceptional detection rates for previously unseen malware variants during validation testing. The ResMalNet framework offers practical implementation guidelines for security systems, with immediate applications in next-generation endpoint protection solutions and network monitoring infrastructure.