Fraud Detection in Decentralized Autonomous Organization (DAO) with Machine Learning

Abstract

A decentralized autonomous organization (DAO) is a type of enterprise that operates on a decentralized structure where all members have equal contribution, right, and decision-making in the organization. This organization makes use of smart contracts, a software where the rules and policies of the organization are embedded. They are susceptible to threats and fraudulent activities which compromise their security. A flaw in the DAO’s smart contract could lead to exploitation by hackers. This research is aimed at developing a fraud detection system using machine learning models and evaluating the performance of the system using standard performance metrics. Machine learning algorithms were employed to detect frauds in the DAO platform built under the Ethereum blockchain using a dataset of transactions, consisting of fraudulent and non-fraudulent transactions. Algorithms employed were the logistic regression, XGBoost, and random forest. These models were built and trained, and hyper parameter tuning was carried out on them. The results obtained from the evaluation metrics show that random forest and XGBoost give better results when compared to logistic regression. Logistic regression had accuracy and precision of 82.07% and 55.53%. Random forest had 70 accuracy and precision of 98.52% and 96.49%. For XGBoost, its accuracy and precision are 98.12% are 93.33%. Other evaluation metrics were used in carrying out analysis, showing the best performing models, the random forest and the XGBoost. At the end of this research, a model for predicting threats in DAO was developed for the ecosystem. The developed system could be utilized in making predictions from past information like patterns of transactions and other available features, to ascertain if the account is fraudulent or not, that account can be terminated.