<![CDATA[The verification process of payment receipts in school environments is still predominantly conducted manually, leading to inefficiency and a high potential for human error. This study proposes an automated system for classifying the validity of digital payment receipts by combining Optical Character Recognition (OCR), BERT (Bidirectional Encoder Representations from Transformers) embeddings, and multi-model machine learning approaches. The system integrates EasyOCR for text extraction from payment receipts, BERT for generating semantic text representations, and four classification algorithms: Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), Naive Bayes (NB), and Logistic Regression (LR). The dataset consists of 185 payment receipt samples, comprising 149 valid and 36 invalid instances, collected via Google Forms and stored in a SQLite database. Experimental results demonstrate that the Multi-Layer Perceptron (MLP) model achieves the highest accuracy of 97% with a test size of 0.2, followed by Logistic Regression with an accuracy of 96.2%, while Naive Bayes exhibits the lowest performance with an accuracy of 85.7%. The proposed system is successfully implemented in a Streamlit-based application, enabling real-time verification of payment receipts with an average processing time of 1.16 seconds per sample.]]>
Copyrights © 2026
