Handwritten Digit Recognition

The use of intelligent systems has become much more common in recent years. Among these, handwritten digit recognition has become a key problem in the field of artificial intelligence and pattern recognition. This study examines the creation and assessment of models for handwritten numerical digit recognition using the MNIST dataset. SVM, MLP, and CNN are the three techniques. We also examine the effects of network width (number of filters per layer), depth (number of convolutional layers), and activation functions (such as ReLU, Tanh, and Sigmoid) on the learning capacity and generalisation ability of the CNN models. CNNs are useful for large-scale handwritten digit classification jobs because, according to experimental data, they provide recognition accuracy and faster processing times than typical machine learning models. In this project their is the use of CNNs and deep learning for handwritten digit recognition. We examine the fundamental architecture of CNNs, which consists of fully connected layers for final classification and decision-making, pooling layers to reduce the dimensions and translational invariance and convolutional layer for feature extraction using filters.