Performance Metric Interpretation on Handwritten Digit Recognition Using Neural Networks

Abstract

In recent trends on up scaling of technology artificial intelligence is much needed one. The AI and ML are the widest field where all the applications and systems are implemented. This paper describes the hand written resignation and specifically to support the medical field. Due to lack of clear writing on prescription sheets, the pharmacologist may deliver the wrong the medicine at sometimes also mostly the scientific papers and land registration documents are become digitalized nowadays. The content to digitalize is high but the method used does not produce reliable values as the error rate is at maximum. Hence, the requirement arises at the research level to simplify the load and reduce the manual efforts, which could done by human. The novelty defines in the system, which can access the prescriptions sheet as image and recognize the hand written on that, and convert them as digitally connected line using different neural network. With respect to the neural network there are many ways to train the dataset and develop the systems which consists of ‘n’ no fo hidden layer. So we prefer implementing the machine algorithm such as SVM, KLN with ANN and CNN. The datasets to train this networks are derived from MNIST database. The train rate are very high. These algorithms compared in terms of reorganization accuracy, and it shows CNN has best of 96.2% compare to other types. The SVM classifier has the better test data accuracy and CNN has the trained data accuracy.