Modified Particle Swarm Optimization for Solving Handwritten Digits

The Handwriting Digits Recognition Problem is a classification problem from interpreting handwritten digit inputs from various sources, such as paper, photos, touch screens, and other devices. This problem is notably challenging and complex due to the uniqueness of everyone’s handwriting, coupled with several crucial characteristics that affect the interpretation process. Many researchers have dedicated their efforts to addressing this challenge. Recent research has made significant strides by enhancing preprocessing and feature extraction techniques, effectively applies with particle swarm optimization to tackle the handwritten digit recognition problem. The results of these experiments have proven to be promising. However, it’s important to note that the particle swarm optimization has its limitations, particularly problem of trapping into local optimum. If we can mitigate or resolve the issue problem of trapping into local optimum within the particle swarm optimization, it has the potential to create superior models. These models can yield higher Recognition Rates during both training and testing phases. Consequently, this paper proposes the incorporation of a mutation technique into the particle swarm optimization. This technique aims to reduce or eliminate problem of trapping into local optimum, ultimately leading to the creation of superior models. When compared to non-mutation techniques, the proposed approach demonstrates higher Recognition Rates when tested with the original MNIST datasets and individual handwritten digit datasets.