Developing Automated Optical Character Recognition System Using Machine Learning Algorithm to Solve Payment Verification Issues

One primary aspect in customer services is to provide immediate solution towards payment verification issues, such as a time delay of payment confirmation by the payment service provider or supplier. This paper presents a development of an accurate optical character recognition (OCR) system using convolutional neural network with deep learning algorithm, which can skip some steps in the workflow of manual payment approval to fasten the process of payment verification and confirmation. By using some machine learning frameworks of pyTorch utilizing Tensors and CUDA-GPU parallel computing, the machine learning based OCR system was developed and tested with the actual data. The real data sets used here cover the non-uniformity of the receipt bill's papers with various conditions (crumple, water drops, and folds) with some nature of the customer's overall camera noise, angle, and lighting. Several experiments associated with data preparation, deep learning parameter settings, and model performance comparison, were properly conducted to obtain a high quality of OCR system to detect trace number, approval codes, and nominals on the widely-used payment receipts. The resulting OCR system performed very satisfactory with 100% accuracy on testing data set. This promising results permit for the integration between this accurate and automated OCR system and chat environment with chatbot technology in order to provide better user experience and immediate and reliable solution toward payment verification issues.