<![CDATA[Manual validation of payment receipts in Enterprise Resource Planning (ERP) applications often faces challenges in terms of Accuracy, especially when payment data must be matched with existing transactions. Data mismatches can lead to recording errors and increase the burden of manual verification. This study aims to improve the Accuracy of payment receipt validation by comparing three Similarity methods: Cosine Similarity, Jaccard Similarity, and Manhattan Distance. In this research, Optical Character Recognition (OCR) is utilized to validate scanned images of payment receipts. By using OCR, data from receipt images can be automatically extracted into text format for further processing. The experimental results show that Cosine Similarity delivers the best performance, with a Precision of 100%, Recall of 90%, and Accuracy of 90%. On the other hand, Jaccard Similarity failed to identify any valid data, resulting in 0% across all evaluation metrics. Meanwhile, Manhattan Distance achieved high Precision (100%) but performed poorly in Recall and Accuracy, both at 10%. Based on these findings, Cosine Similarity is recommended as the most effective method for enhancing OCR-based payment validation in ERP systems. This study also opens the opportunity to develop hybrid approaches, combining Cosine Similarity and Manhattan Distance methods to further improve overall system performance.]]>
Copyrights © 2025
