Vulnerability detection in smart contracts using quantum machine learning

The rapid adoption of blockchain technology has brought significant concerns, particularly regarding the security and integrity of smart contracts. This study focuses on identifying potential vulnerabilities in smart contracts, emphasizing that such weaknesses can lead to serious financial losses and undermine trust in blockchain systems. In this context, a Hybrid Quantum-Classical Deep Neural Network (HQCDNN) model that integrates the fundamental principles of quantum computing with classical deep learning techniques has been proposed. The proposed model is designed to detect common security vulnerabilities such as access control, arithmetic, front-running, reentrancy, time manipulation, denial of service, and unchecked low calls. During the model training process, the SmartBugs Wild dataset was utilized, and the Term Frequency-Inverse Document Frequency (TF-IDF) method optimized to suit hybrid architectures was employed for feature extraction. In the experimental studies, hybrid architectures incorporating both 2-qubit and 4-qubit quantum layers were designed and tested. To enable comparative analysis, a conventional Deep Neural Network (DNN) model was also included in the process. The HQCDNN model demonstrated superior performance compared to classical and deep learning-based methods, achieving accuracy rates ranging from 96.4% to 78.2% and F1 scores ranging from 96.6% to 80.2%. These results indicate that hybrid quantum classical models offer high accuracy and efficiency, particularly in the field of smart contract security. Future studies may aim to test this model on real quantum hardware and conduct comprehensive evaluations using larger and more diverse datasets.