A Framework for Evaluating Performance in Fake Currency Detection Utilizing Machine Learning Models

Counterfeit banknote circulation continues to pose a serious threat to financial stability worldwide. Manual inspection methods are unreliable and increasingly circumvented by high-precision forgeries. This paper presents a comprehensive evaluation framework that applies seven supervised machine learning (SML) algorithms—K-Nearest Neighbor (KNN), Naïve Bayes (NB), Decision Tree (DT), Support Vector Machine (SVM), Random Forest (RF), Logistic Regression (LR), and the extended LightGBM algorithm—to the UCI Banknote Authentication dataset. Experiments are conducted across three train test split ratios (80:20, 70:30, and 60:40) and models are evaluated using Accuracy, Precision, Recall, F1-Score, and Matthews Correlation Coefficient (MCC). LightGBM achieves 100% accuracy under the 80:20 split, outperforming all traditional classifiers. The results demonstrate that gradient-boosted ensemble methods provide a reliable, automated solution suitable for integration into banking and ATM infrastructure. Keywords—Fake Currency Detection, Banknote Authentication, Supervised Machine Learning, LightGBM, Random Forest, SVM, KNN, Naïve Bayes, Decision Tree, Logistic Regression, UCI Dataset, Feature Extraction, Financial Security, Classification