Counterfeit currency presents a substantial risk to economic stability and financial security, necessitating efficient and dependable detection techniques in both forensic and practical contexts. This research examines digital forensic methodologies for the identification of counterfeit banknotes employing three machine learning classifiers: K-Nearest Neighbors (KNN), Logistic Regression, and Support Vector Machine (SVM). A dataset was generated by photographing authentic and counterfeit Indonesian banknotes using a mobile phone camera, thereafter undergoing preprocessing and augmentation to enhance resilience. To improve classification performance, three image preprocessing techniques—grayscale filtering, edge detection, and blurring—were employed. The models were assessed based on accuracy, precision, recall, and F1-score obtained from confusion matrix analysis. The experimental findings demonstrated that SVM and Logistic Regression consistently surpassed KNN in all settings, with SVM attaining the best overall accuracy of 0.997 under gray and blur filtering. Logistic Regression exhibited high reliability, with an accuracy of 0.994–0.997 using gray and blur filters. KNN, although originally less successful, showed significant enhancement when integrated with blur filtering, attaining an accuracy of 0.973. Conversely, edge detection was found to be detrimental to the performance of all tested models.