Digit Recognition using Singular Value Decomposition (SVD)

Overview

This project focuses on digit recognition using Singular Value Decomposition (SVD) as a dimensionality reduction technique. The dataset used for training and testing contains handwritten digits in a matrix format.

Prerequisites

Make sure you have the following dependencies installed:

• Python (>=3.6)
• Required Python packages: matplotlib, pandas, numpy, scipy, scikit-learn

Dataset

The dataset consists of two main parts:

• Training Data:
  • Features (xtrain): Handwritten digit images in matrix form.
  • Labels (ytrain): Corresponding digit labels for training.
• Testing Data:
  • Features (xtest): Handwritten digit images for testing.
  • Labels (ytest): True labels for testing data.

Preprocessing

1. Displaying Digits: The show_digit function takes the index of a column and displays the corresponding digit from the dataset.
2. Creating Matrices: Matrices for each digit (0-9) are extracted from the training data.

Dimensionality Reduction with SVD

1. SVD Analysis: Singular Value Decomposition is applied to each digit's matrix.
2. Optimal Number of Basis Vectors: The accuracy of the digit recognition model is evaluated for different numbers of basis vectors.

Results

1. Accuracy Scores: The accuracy scores for various numbers of basis vectors are displayed.
2. Confusion Matrix: A confusion matrix is generated for the model with the optimal number of basis vectors.

Observations

1. Common Misclassifications: Some commonly misclassified digits are displayed, providing insights into the model's performance.
2. Impact of Changes: The impact of changing the number of basis vectors for specific digits is explored.