【Author】
Gupta, Rajesh; Patel, Mohil Maheshkumar; Shukla, Arpit; Tanwar, Sudeep
【Source】COMPUTERS & ELECTRICAL ENGINEERING
【Abstract】Smart contracts are essential in maintaining the trust between the members of the blockchain. Its verification is of utmost importance, as it is unmodifiable once deployed. Moreover, a malicious user can deploy vulnerable smart contracts to breach the blockchain data. To restrict this, we propose a deep learning-based scheme to detect the vulnerabilities and rate them as safe/vulnerable based on probability value < 0.5/>= 0.5 respectively. An open Google BigQuery dataset with 7000 samples was used to train the classifier. We train artificial neural networks (ANN), long-short term memory (LSTM), and gated recurrent unit models (GRU) and compare their accuracy, precision, recall, and receiver operating characteristic (ROC) curve values. Results show the LSTM model outperforms ANN and GRU. Then, we simulate the LSTM to classify the smart contracts before their deployment in the blockchain. Also, the efficacy of the blockchain is justified with the proposed system's data storage cost and scalability.
【Keywords】Deep learning; Smart contract; Blockchain; Internet of things; LSTM; GRU; ANN; Malicious
评论