Enhanced Fraud Detection in Ethereum Transactions: Fusion of Modified Genetic Algorithms and Deep Learning with Limited Attributes

Abstract

Ethereum smart contracts, the new way of transactions and a popular name in the world of cryptocurrencies, have gathered a huge base of research and scientific attention. They are so helpful that they allow us to eliminate the need for a separate third-party library to allow unknown parties to see contract details on a computer. But since we can all see that online commerce is growing day by day and will continue to grow, it can never be fully free from scams and unethical operations. So, to correctly detect all such unethical and malicious transactions, this paper used a deep learning model. And to further enhance the model, this study used metaheuristic optimization as well. It employs an algorithm called Genetic Algorithm and to provide better optimization in the explorations phase of Cuckoo Search (CS) to achieve its goal of detecting fraudulent transactions. The algorithm covers the loopholes in the CS strategy. Furthermore, to provide strong grounds for research in our paper, this proposed model was compared with various types of approaches such as Light Gradient Boosting Machine, Support Vector Classification, Multi-Layer Perceptron, XGBoost, Logistic Regression, and Random Forest. The proposed model outperforms other significant models like SVC, KNN, LGBM Classifier, RF etc. with the accuracy of $\mathbf{9 8. 6 \%}$.