Optimizer Based Performance Study of CNN Architecture for MODI Handwritten Character Recognition

The script known as MODI, originated in Maharashtra during the rule of Chhatrapati Shivaji Maharaj, widely used to create official documents. To make MODI recognition easier, a recognition system for handwritten characters might be established. Character recognition is one of the pattern identification applications that use deep learning-based algorithms. In this research, we explored a deep learning Convolutional Neural Network (CNN) model for character recognition based on essential performance parameters such as learning rate, batch size, and optimizer. For the performance evaluation of CNN, optimization methods such as Adam, Sgdm, and RMSProp on batch sizes of 8, 16, 32, and 64 with learning rates of 0.01, 0.001, and 0.001 are examined, respectively. Our CNN architecture is implemented on a self-created dataset consisting of 11,200 images of MODI characters to attain recognition accuracy of handwritten characters. In accordance with the test results, using the Adam optimization with a learning rate of 0.0001 and a batch size of 32, a prediction accuracy of 99.96% is acquired. Thus, this result may be helpful in recognizing various scripts in the future.