Detection of Digital Currency Fraud through a Distributed Database Approach and Machine Learning Model

The world is witnessing a noticeable increase in financial exchange in digital currencies such as Bitcoin, Ethereum, and others, as transactions in electronic markets have begun to rise recently, which increases the difficulty of maintaining security and trust in decentralized financial systems that use distributed databases and the technologies that interact with them in Ethereum networks, blockchain, etc. This study presents a hybrid model based on the PyCaret library and includes 12 machine learning classifiers, with the aim of identifying fraudulent activities in Bitcoin transactions and enhancing the security of Ethereum networks and blockchain technology. The results reveal the effectiveness of different models in identifying fraudulent activities on the Ethereum network through a comprehensive performance comparison. The classifiers that showed the highest accuracy scores, which ranged from 0.9814 to 0.9862, were the Random Forest classifier, the visual gradient boosting machine, and the additive tree classifier. It is important to note that both Gradient Boosting Classifier and K Neighbors Classifier performed well, with accuracies above 0.96 and AUC scores above 0.99. However, some models, such as Naive Bayes, showed lower accuracy and AUC scores, suggesting that they have limitations in terms of accurately detecting fraudulent transactions. These results highlight the importance of choosing appropriate machine learning models for fraud detection tasks in general, with ensemble techniques such as Extra Trees and Random Forest showing great promise in this regard.