Modified Feature Selection with BPSO to Apply PSO for Solving Handwritten Digits

The task of Handwriting Digits Recognition represents a classification challenge involving the interpretation of handwritten digits sourced from diverse mediums such as paper, photos, touch screens, and other devices. This challenge is very intricate due to the distinctiveness of each individual's handwriting, and several essential traits that influence the interpretation process. Numerous researchers have devoted efforts to solve this problem. Recent research has improved preprocessing and feature extraction techniques by using particle swarm optimization to address the handwritten digit recognition problem. The outcomes of these experiments have shown obtain good results. Nevertheless, outcomes from this technique can be further improved. The feature selection technique popularly applies pattern recognition to improve outcomes. Thus, this study proposes a modified feature selection technique that developed from binary particle swarm optimization to optimize particle swarm optimization for handwritten digit recognition. This proposed feature selection technique has demonstrated the potential to yield higher Recognition Rates significantly in both training and testing phases. Comparative analysis against the original technique, devoid of the feature selection technique, reveals that the proposed method exhibits superior Recognition Rates when tested on both the original MNIST datasets and the individual's each person handwritten digit datasets.