Design and analysis of LRTB feature based classifier applied to handwritten Devnagari characters: A neural network approach

This research work investigates design and robustness analysis of an optimal classifier applied to handwritten Devanagari consonant characters using single hidden layer feed-forward neural network with respect to five fold cross validation. Proposed neural network is trained 3 times by varying PEs in hidden layer from 64 to 128 in steps of 16. For each fold fifteen neural networks are studied. Meticulous experimentation of around seventy five MLPs shows the overall classification accuracy near to 97% for all classes. The best network is found at fold 5 with 80 neurons at trial 3. Networks analyzed on account of confusion matrix, reveals the greater details for individual classes. Average classification accuracy on training, validation, test and combined dataset is 99.40%, 97.38%, 97.05% and 98.98% respectively on the total dataset size of 8224 samples distributed uniformly within 32 classes of typical Devnagari consonants.