Analysing Vulnerabilities in Smart Contracts Using Source Code and Transaction Level Features

Abstract

Smart contracts are self-executing and tamper-resistant programs deployed on blockchain networks. Their decentralized and autonomous nature enables various industries to build applications that rely on a single source of truth. However, the pseudonymous nature of blockchain, combined with its immutable design, presents a complex landscape for regulation and accountability. While immutability enhances transparency and traceability, it may also hinder the reversal of malicious or fraudulent transactions, potentially facilitating certain illegal activities. This research compares the contributions of source code and transaction metadata-based features in vulnerability detection. Using the data analysis techniques to correlate vulnerabilities with specified features. The proposed approach achieved nearly 83% accuracy for the source code domain and 55% for the transaction domain. The validation was done by comparing the performance against machine learning models. The results showed that code-based features expose the vulnerabilities of smart contracts more effectively than transaction metadata-based features.