Improved F-Measures Over Gaussian SVM for Handwritten Digit Recognition Using LSTM to Increase Accuracy

This study focuses on creating a model to recognize handwritten digits using LSTM-based networks and comparing its performance with a Gaussian SVM for optical character recognition. We used MNIST dataset, which contains <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{7 0, 0 0 0}$</tex> handwritten digit images. Out of these, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{6 0, 0 0 0}$</tex> images were used for training the models, and 10,000 were used for testing. To ensure our sample size was adequate, we used G-Power statistics software to conduct an 80 % power analysis, setting the alpha error rate at 0.05 to minimize type-I errors. The comparison showed that the Gaussian SVM model achieved an accuracy of 88 % in recognizing the digits. In contrast, the LSTM model performed much better, reaching an impressive 99% accuracy. The difference in performance was statistically significant, with a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{p}$</tex>-value of less than <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$0.0001(p&lt;0.05)$</tex>. These findings indicate that LSTM networks significantly outperform Gaussian SVMs in recognizing handwritten digits, making them a more effective choice for optical character recognition tasks. This research highlights the advantages of LSTM over traditional SVM approaches in handling complex recognition tasks.