Neural network based approach for Ethereum fraud detection

Abstract

Over the years with the advent of technology 4.0, there have been an increased inclination towards the blockchain technology and the cryptocurrencies. Thus it is becoming imperative to implement fraud detection techniques over the Ethereum network which nowadays is a popular platform for the developers to create smart contracts and deploy decentralized apps. Over the years the various machine learning algorithms have been growing for the process of fraud detection and in particular neural networks have shown promising results. As such in this paper, a neural-network based approach has been presented for Ethereum fraud detection and to validate the effects of the performance this proposed model has been compared with its peers. In contrast to the various models such as Logistic Regression, SVM, Gaussian Naive Bayes, K-nearest neighbour, the neural network perform the best providing an accuracy of about 97.09% which is higher than the rest. It is then seen that neural networks are relatively effective in learning complex patterns of the dataset and thus classifying the resultant transaction as genuine or fraudulent. Thus this work contribute in the development of effective solutions for fraud detection in the Ethereum and other blockchain platforms, enhancing their security and reliability.