MNIST Handwritten Digit Recognition

CNN-based digit classifier achieving 99% accuracy on MNIST dataset.

Model Architecture

• 3 Conv2D layers (64 filters each, 3x3 kernel)
• 2 MaxPooling layers (2x2)
• Dropout (0.5) for regularization
• Dense layer with softmax (10 classes)

Quick Start

bash pip install tensorflow numpy matplotlib python from keras.datasets import mnist from keras.utils import to_categorical from keras import layers, models

Load and preprocess data

(train_images, train_labels), (test_images, test_labels) = mnist.load_data() train_images_processed = train_images.reshape((60000, 28, 28, 1)).astype('float32') / 255 train_labels_categorical = to_categorical(train_labels)

Build model

model = models.Sequential([ layers.Conv2D(64, (3, 3), activation='relu', input_shape=(28, 28, 1)), layers.MaxPooling2D((2, 2)), layers.Conv2D(64, (3, 3), activation='relu'), layers.MaxPooling2D((2, 2)), layers.Conv2D(64, (3, 3), activation='relu'), layers.Flatten(), layers.Dropout(0.5), layers.Dense(10, activation='softmax') ])

model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy']) model.fit(train_images_processed, train_labels_categorical, epochs=5, batch_size=64) Making Predictions python

Predict single image

def predict_image(index): pred = model.predict(test_images_processed[index:index+1]) digit = np.argmax(pred) confidence = np.max(pred) * 100 print(f"Predicted: {digit} | Actual: {test_labels[index]} | Confidence: {confidence:.2f}%") return digit

Predict first 10 test images

predictions = model.predict(test_images_processed[:10]) for i in range(10): print(f"Image {i+1}: Predicted = {np.argmax(predictions[i])}, Actual = {test_labels[i]}")

Results

Test Accuracy: 99.2% Training: 5 epochs, 64 batch size

Loss: 0.0254

Key Functions

Function Purpose

• predict_image(index) - Predict digit by test image index
• model.predict() - Batch predictions
• np.argmax() - Get predicted digit from probabilities

Requirements

• text
• tensorflow>=2.0.0
• numpy>=1.18.0
• matplotlib>=3.0.0

Author

Collins Kimani - @collinskimani482-sudo