Handwritten Digit Recognition using K-Nearest Neighbors (KNN)

Description

This project implements a Handwritten Digit Recognition system using the K-Nearest Neighbors (KNN) algorithm. The system uses the MNIST dataset, a widely-used dataset containing 70,000 grayscale images of handwritten digits (0-9), each 28x28 pixels in size. The model is trained to classify these digits with high accuracy.

The code also includes a user-friendly functionality to predict handwritten digits from user-provided image files. The user can provide an image, and the model will preprocess and classify the digit based on the trained KNN model.

---

Installation

1. Clone the repository:

   git clone https://github.com/your-username/handwritten-digit-recognition.git
   cd handwritten-digit-recognition

2. Install dependencies: Make sure you have Python 3.7+ installed. Install the required Python libraries by running:

   pip install numpy
   pip install pandas
   pip install matplotlib
   pip install Pillow
   pip install scikit-learn

3. Dataset download: The MNIST dataset is automatically fetched using the fetch_openml function from scikit-learn.

---

Usage

Train and Evaluate the Model

1. Run the main script to train the KNN model, evaluate its accuracy, and determine the best value of k using the elbow method:

   python main.py

2. The script will:

   • Load the MNIST dataset.
   • Split it into training and testing sets.
   • Find the optimal value of k using the elbow method.
   • Train the KNN classifier with the optimal k.
   • Evaluate the model using metrics such as accuracy and confusion matrix.

Predict Custom Images

1. Prepare your custom handwritten digit image:

   • Ensure the image is grayscale.
   • Save the image in a file format like .png or .jpg.

2. Run the script and provide the image path when prompted:

   python main.py

   • Enter the path to your image file when prompted.

3. The script will preprocess the image and output the predicted digit.

---

Example

Output after training and testing:

• Optimal k: The elbow method determines the best value for k, for example: k = 1.
• Accuracy: The script outputs the accuracy of the model, e.g., Accuracy with k = 1: 97.06%.
• Confusion Matrix: A confusion matrix is displayed to show the classification performance.

Output for a custom image:

• After preprocessing and prediction:

  Enter the path to image: path/to/your/image.png
  Predicted digit: 5
  

---

Notes

• Ensure your custom images are clear and have a resolution similar to the MNIST dataset (28x28).
• Modify the script if needed to experiment with different values of k or preprocessing steps.
• For any questions or issues, feel free to open an issue in the repository.