Handwritten Digit Recognition Using Convolutional Neural Networks (cnn) On The mnist dataset

Abstract

Handwritten Digit Recognition Is Crucial In Various Automated Tasks Like Financial Verification, Postal Sorting, And Digital Form Processing, Where Correctly Understanding Handwritten Numbers Is Key. This Study Looks At How Well Convolutional Neural Networks (CNNs) Work For Recognizing Handwritten Digits, Using The MNIST Dataset, Which Is A Common Test Set Made Up Of Black And White Images Of Numbers. The Research Follows A Structured Approach That Includes Normalizing The Dataset, Reshaping Images, And Building A Multi-layer CNN Structure Designed For Classification. The Proposed Model Uses Three Convolutional Layers With Increasing Numbers Of Filters, Along With Max-pooling, Dropout Regularization, And Fully Connected Dense Layers To Capture And Understand Important Visual Patterns. The Network Is Trained Using Categorical Cross-entropy Loss And Stochastic Gradient Descent Over 15 Training Cycles. The Experimental Results Show The Model Achieves A Test Accuracy Of 99.2%, Which Shows Its Strong Ability To Generalize And Perform Well Across Different Handwriting Styles. A Confusion Matrix Analysis Further Confirms Consistent Performance Across All Digit Classes, With Very Few Misclassifications. These Results Highlight The Potential Of CNN-based Models For Reliable Visual Recognition And Lay The Groundwork For Future Work With More Complex Datasets And Expanded Character Recognition Systems. The Study Shows That Well-designed CNN Structures Can Achieve Very Reliable Recognition Of Handwritten Digits, Providing A Solid Base For Future Progress In Automated Visual Understanding Systems.

Keywords