Hand-written Digits Classification using TensorFlow

An end-to-end Deep Learning project that classifies handwritten digits (0-9) from the famous MNIST dataset using a Multi-Layer Perceptron (MLP) built with TensorFlow and Keras.

---

🚀 Project Highlights

• High Accuracy: Reached over 97.83% accuracy on the test dataset.
• Balanced Architecture: Optimized using a Dropout layer to robustly eliminate overfitting.
• Production-Ready: Saves the trained model pipeline into a reusable .keras format.
• Verification Built-In: Performs automated script tests comparing real sample labels directly with inference predictions.

---

📊 Dataset Structure

The MNIST dataset is automatically loaded and handled within the script:

• Training Set: 60,000 images ($28 \times 28$ pixels)
• Test Set: 10,000 images ($28 \times 28$ pixels)
• Features Scaling: Pixel values normalized from [0, 255] to [0.0, 1.0] for optimal gradient descent stability.

---

🧠 Neural Network Architecture

The model is sequentially structured as follows:

1. Flatten Layer: Reshapes the 2D image matrix ($28 \times 28$) into a flat 1D array of 784 pixels.
2. Dense Hidden Layer: 128 neurons utilizing the ReLU activation function.
3. Dropout Layer: Randomly drops out 20% of nodes during training epochs to boost generalization.
4. Dense Output Layer: 10 neurons mapping out multi-class probability scores via Softmax.