Advanced Phishing Detection in Ethereum Blockchain Transactions Using Machine Learning Models

Abstract

Deceptive phishing attacks greatly endanger blockchain security, tricking miners into adding harmful blocks to the chain. Current methods of detection and agreement protocols are frequently not enough, especially if authorized miners accidentally include these blocks. Despite the potential for improving detection capabilities, the adoption of zero-trust policies is still restricted. This paper explores different machine learning techniques, like k-Nearest Neighbors (k-NN), Decision Trees (DT), Random Forest (RF), and XGBoost, to predict phishing attacks. It also evaluates feature selection methods such as Principal Component Analysis (PCA) and Decision Trees, ultimately recommending the Random Forest (RF) model as the most effective for phishing detection. The RF model, assessed using metrics such as accuracy, precision, recall, and evaluation time, demonstrates superior performance, achieving up to 99% accuracy. Consequently, the RF model emerges as the optimal choice for accurately and efficiently identifying phishing threats, thereby enhancing the security of blockchain networks.