ADAPTIVE MACHINE LEARNING TECHNIQUES FOR REAL-TIME FRAUD DETECTION IN BANK TRANSACTIONS

Abstract

The rapid growth of blockchain-based financial transactions and digital banking services has increased the need for reliable fraud detection mechanisms. Although blockchain consensus verifies transaction validity, it cannot independently identify malicious participants or fraudulent behavioural patterns. This paper presents a comparative machine learning framework for real-time fraud detection in blockchain transaction environments. The proposed methodology uses Node2Vec for network embedding and feature extraction, followed by normalization and bootstrap sampling. Multiple supervised algorithms, including Logistic Regression, Naive Bayes, AdaBoost, Decision Tree, SVM, Random Forest, MLP, and Neural Network, are trained and evaluated. Performance is measured using Accuracy, Precision, Recall, F1- Score, ROC-AUC, Confusion Matrix, and computational efficiency. Results indicate that ensemble methods, particularly Random Forest and AdaBoost, achieve accuracy above 97%. The framework provides an additional verification layer that improves blockchain transaction security without modifying the underlying consensus mechanism. Keywords: Blockchain Fraud Detection, Machine Learning, Random Forest, AdaBoost, SVM, Financial Security.