Quantum Inspired Feature Selection for Efficient Early Ransomware Detection

Abstract

The increasing prevalence and sophistication of ransomware pose a significant challenge to cybersecurity. Traditional mitigation methods often fail to detect threats during early stages, especially when malware presents limited behavioral signatures. This study introduces a hybrid feature selection framework that integrates classical machine learning models with a Quantum Unconstrained Binary Optimization (QUBO)-based formulation to identify early-stage ransomware indicators. Our approach evaluates the relevance and redundancy of static features extracted from Windows Portable Executable (PE) files. By solving the QUBO problem using a simulated annealing algorithm, we achieve a compact feature subset that preserves predictive performance. Comparative experiments involving XGBoost, Random Forest, and other classifiers demonstrate that the QUBO-selected subset, consisting of only six features, performs comparably to full feature sets, with an F1 score exceeding 0.99. These findings highlight the potential of quantum-inspired techniques for lightweight malware detection on resource-constrained systems.