Project-2

🖊️ KNN on MNIST – Handwritten Digit Recognition

This project implements the K-Nearest Neighbors (KNN) algorithm on the famous MNIST handwritten digits dataset.
It includes accuracy tuning, performance evaluation visualizations, and even a simple GUI where you can draw a digit and see the model's prediction in real time.

---

📌 How It Works

KNN is a simple yet powerful algorithm:

1. For each new image, it finds the k most similar images in the training set.
2. These neighbors "vote" for the most common digit among them.
3. The digit with the most votes is the predicted label.

In this project:

• Accuracy vs k was plotted to choose the best k value (k=3 gave the highest validation accuracy).
• Confusion matrix and classification report were generated to analyze the model's performance.
• A GUI was created to test the model with your own handwritten digits.

---

📊 Results

• Best k: 3
• Highest validation accuracy: ~97.1%
• Final test accuracy: 94.5%
• Strong performance across most digits, with occasional confusions between visually similar digits (like 4 and 9).

---

📷 Visuals

Accuracy vs k

Normalized Confusion Matrix

Classification Report

Sample Predictions

---

🎨 GUI

A simple graphical interface is included to draw digits and get real-time predictions from the KNN model.
Run the GUI using:

python knn_gui.py