Abstract
HAND WRITING RECOGNITION SYSTEM FOR FRAUD DETECTION USING DEEP NEURAL NETWORK
Nkolika O. Nwazor* and Babangida Bakoji
ABSTRACT
Financial institutions are faced with the menace of fraud committed by criminals who forge the signatures of their customers on bank documents. This work is on the development of a handwritten text recognition system using deep neural network. Offline handwritten text recognition technique was used in this work. The code for the model was written in Keras, a backend for TensorFlow used typically as an application programmable interface (API) for building deep learning model. The system was trained using the handwritten text of one of the authors. The performance of the training model was improved by changing the learning rate, the dropout threshold, the batch size, and the number of iterations (hyper parameter tuning). The accuracy of the system in predicting handwritten text was found to be 0.9806 i.e. 98.06%. An application was developed for the fraud detection. The system was evaluated using a written text it was not trained with and it was able to detect it as a fraud.
[Full Text Article] [Download Certificate]