Predictive Analytics for Fake Currency Detection

Abstract

Fake currency is the imitation of the real currency without the legal sanction of the government. Producing or using fake currency is a form of forgery and it is illegal. The circulation of fake currency increases inflation and raises demands for goods which also affects the producers and consumers thereby leads to currency devaluation. Classifying fake currency on the fly is a major challenge. The aim of this work is to detect fake currencies using predictive analytic models by considering the statistical characteristics of the fake currency note as features. The US dollar images are considered for classification of fake currency notes. The statistical characteristics variance, skewness, kurtosis and entropy of the fake currency images are extracted from images and are treated as the independent variables for model building. The dataset is splitted into 80:20 as training and test datasets. The classifiers Naïve Bayes, Random Forest, AdaBoost, Logistic Regression, Support Vector Machine, K- Nearest Neighbor and Multi-layer Perceptron are built. The performance metrices accuracy, precision, recall and F1 score of the built models are compared. The predictive analytics models SVM, KNN and MLP yield the best accuracy 100% and these models are not suffering from drift between the train and test dataset.