Fake Currency Detection Across Five Classes: A CNN-SVM Based Approach
Shiva Mehta, Anubhav Bhalla · 2024
The presented work offers an integrated model combining a Convolutional Neural Network (CNN) to extract features and Support Vector Machines (SVM) for labelling counterfeit Cash. The Model is intended to identify the counterpart of actual currency in ten types of notes, including five genuine and five of the four kinds of counterfeit notes. During model creation, it was found that the system achieved an overall accuracy of 84%. The survey revealed that 82% of the engaged population were using a database of 2600 photographs. The study of a confusion matrix demonstrates a high level of performance, particularly for class 1, which reaches 80 accuracies as a result. There is a 69% print run, and the listens should be 97% ROC Curve display the acute discriminative capability, as the AUC values attain 0.74. Top four percentiles have Class 0, 0.85 in Class 1 and 0 for difference level teachers-75 for Class 2. The error graph in training and validation illustrated the learning curve, which was a good performer as the error consistently grew over the 200 epoch iterations. The follow-up assessment with bar graph representation showed an 84% achievement average. Overall, the fault reported in the ground vehicle is 77%, and the micrometric recall average is 86.09%, with a macro F1-score of 85 being the average. Of all the counts, 02%. The precedent matches 85% Power of Precision. 36%, recall of 84. It achieved a recall and the F1-score of 82% and 84%. 91% of the micro metrics were retained at 84 precious memories, and F1-Score is 82%.