Feature Selection Approaches in Online Bangla Handwriting Recognition

Any feature selection technique aims to identify a smaller subset of essential characteristics from a larger collection by eliminating those that are redundant, noisy, or irrelevant. Feature selection techniques have proven to be a major ground-breaking technique to save computing time while strengthening prediction accuracy and data interpretation and better cognition of data in machine learning as well as pattern recognition algorithms. In the present experiment, we have compared four well-known feature selection techniques, Harmony Search, Krill Herd Algorithm, Principal Component Analysis and mRMR Algorithm when applied on the feature set produced by combining the frechet distance and distance-based features (256-dimension feature vector). The experiment has been performed on a 10000 online handwritten Bangla character database. The experimental outcome metrics and the analysis of the performances with respect to the percentage of dimensionality reduction and achieved accuracy are presented in a nutshell.