| Autor: |
J, Lohith J, K, Anusree Manoj, P, Guru Nanma, Srinivasan, Pooja |
| Zdroj: |
Multimedia Tools & Applications; Sep2023, Vol. 82 Issue 23, p36379-36393, 15p |
| Abstrakt: |
Smart contracts are a set of instructions or programs that are stored on the blockchain network and run when predetermined conditions are met. Ethereum smart contracts are deployed on blockchain networks. Ethereum smart contracts are immutable and are vulnerable to simple coding errors called vulnerabilities. The aim of this paper is to classify Ethereum smart contracts vulnerabilities by feature extraction using machine learning. Pixel values collected from images and trigram feature extraction were used to construct the dataset. This dataset was trained using Multilabel k Nearest Neighbours (MLkNN), Binary Relevance kNN (BRkNN), Random Forest (RF), and Naive Bayes (NB), among other machine learning methods. The Naive Bayes Method outperforms the other models in terms of F1-score among all the algorithms tested. The Naive Bayes model achieves F1-scores of 99.38% and 99.44% using Binary Relevance and Classifier Chain respectively. In terms of F1-score, the Random Forest model attained a substantial degree of performance, with F1-scores of 96.71% and 96.61% using Binary Relevance and Classifier Chain respectively. In comparison, the lazy algorithms MLkNN and BRkNN produced lower F1-scores of 88.19% and 89.71%, respectively. This suggests that using the TriPix dataset outperforms models employed in either opcode characteristics or image-based detection used in other works. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
Full text from SpringerLink