Predicting and Mitigating Fraud and Money Laundering Using Machine Learning Models: Logistic Regression, Support Vector Machines, Decision Trees and KNN

The global financial system faces persistent risks posed by money laundering, necessitating sophisticated and timely detection mechanisms beyond conventional rule-based systems. This study addresses this critical challenge by performing a comparative analysis of four distinct machine learning algorithms: Logistic Regression (LR), K-Nearest Neighbors (KNN), Support Vector Machines (SVM), and Decision Trees (DT). The primary objective was to evaluate the efficacy and generalization capability of these models in classifying suspicious transactions with high accuracy. The dataset, comprising various financial features, was evaluated using performance metrics derived from both training and unseen test datasets.

The results demonstrated significant variations in model performance. While Logistic Regression achieved a modest classification score (63.0% on both training and testing sets), KNN (84.0% test score) showed improvement over the baseline. Both SVM (85.0% test score) and Decision Trees (86.0% test score) significantly outperformed their counterparts, exhibiting superior predictive power and robust generalization. Specifically, the Decision Tree Classifier achieved the highest performance consistency, reporting both an 86.0% training score and an 86.0% test score.

The findings strongly suggest that complex non-linear models are essential for effective AML prediction. The observed superiority of Decision Trees indicates its potential to capture intricate relationships within the financial data better than linear or simple proximity methods. These results provide empirical evidence recommending advanced classification techniques, such as Decision Trees, for deployment in real-world Anti-Money Laundering systems, thereby enhancing institutional fraud detection capabilities and contributing to financial integrity.