A Handwritten Digit Recognition using with Convolutional Neural Networks with Compared K-Nearest Neighbor with Improved Accuracy

The goal of this project is to enhance and develop a system for identifying handwritten digits by addressing key issues in pattern recognition. We compare two different approaches, Convolutional Neural Networks and Decision Trees. In our study, we applied the Convolutional Neural Networks algorithm to a dataset of handwritten digits with a sample size of 66. We compared this with the K-Nearest Neighbors (KNN) algorithm, also tested with a sample size of 66. For training, we used 80 % of the data and for testing, we split the remaining data into 70 % for testing and 30 % for validation. This approach helped us gauge how well each algorithm performs. The results showed that the CNN classifier achieved an impressive accuracy rate of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 6. 4 2 \%}$</tex>, while the KNN algorithm reached <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 1. 2 1 \%}$</tex>. This significant difference in accuracy was statistically validated with a p-value of 0.001, indicating a strong contrast between the two methods. The confidence interval for these results is 95 %, further confirming the reliability of the findings.