Análise de defeitos em aplicativos descentralizados no contexto da engenharia de software orientada a blockchain

Abstract

The analysis of defects in decentralized applications (DApps) within the context of blockchain-oriented software engineering is crucial due to the autonomous and self executing nature of these programs, which allow for the execution of agreements without intermediaries. However, like any software, DApps are susceptible to defects and can present vulnerabilities exploitable by attackers. The use of models for software defect prediction is a well-studied research area, but the application of these models with smart contract metrics is still underexplored. This thesis project aims to evaluate whether Deep Learning models used in traditional software defect prediction produce equivalent results when applied to specific smart contract metrics. To this end, the research proposes the use of a deep neural network called Deep Neural Network—Ethereum Solidity Metrics (DNN-ESM). The methodology includes data preparation, the selection of machine lear ning models, and performance evaluation using metrics such as precision, recall, F1-Score, area under the curve (AUC), precision-recall curve (PRC), and Matthews correlation coef f icient (MCC). The machine learning models will be applied to datasets containing traditi onal object-oriented software metrics and specific metrics for smart contracts in Solidity. The results suggest that deep learning models can be effective in predicting defects in smart contracts, improving efficiency and reducing the computational effort required for vulnerability detection. Furthermore, the application of these models will allow for the evaluation of a large number of smart contracts in a short period, benefiting developers, researchers, and educators in the field of blockchain-oriented software engineering.