MNIST Digit Classification – README

This project builds a complete machine learning workflow for classifying handwritten digits from the MNIST dataset using multiple models and ensemble learning techniques.

---

📌 Project Overview

This notebook performs:

• MNIST data fetching
• Visualization of sample digits
• Class distribution analysis
• Feature scaling
• Training/test splitting
• Evaluation of multiple classification algorithms
• Cross-validation using StratifiedKFold
• Ensemble learning via Voting Classifier
• Model comparison using accuracy boxplots

---

📁 Dataset: MNIST

The MNIST dataset consists of 70,000 grayscale images of handwritten digits (0–9), each sized 28×28 pixels.

Each sample has:

• 784 pixel values (flattened 28×28 image)
• 1 label (digit 0–9)

Loaded using:

mnist = fetch_openml('mnist_784', version=1)

---

🧭 Workflow Summary

1. Import Required Packages

Includes NumPy, Pandas, Seaborn, Matplotlib, Scikit-Learn, and XGBoost.

2. Load and Inspect the Dataset

• Check structure and keys
• Display image samples
• Show class distribution (balanced dataset)

3. Visualizations

• Display single digit image
• Grid of 30 sample images using a custom print_image() function
• Countplot for digit class frequencies

4. Split the Dataset

To speed up training:

• Take 50% of MNIST → X_small, y_small
• Train/test split on reduced dataset

X_train, X_test, y_train, y_test = train_test_split(...)

---

⚙️ Preprocessing

🔹 Feature Scaling

Standardization using:

StandardScaler()

This helps gradient-based and distance-based classifiers.

---

🧠 Models Used

This project evaluates multiple classifiers:

Individual Models

• Logistic Regression
• Gaussian Naive Bayes
• Random Forest Classifier
• Gradient Boosting Classifier
• K-Nearest Neighbors (KNN)

Ensemble Model

A Voting Classifier combining:

• Logistic Regression
• Random Forest
• Gradient Boosting

Uses soft voting for improved performance.

---

📝 Model Evaluation

🔹 Cross-Validation

Performed using StratifiedKFold (5 splits) to maintain class balance. Scores are calculated using accuracy metric.

🔹 Test Set Evaluation

Each model prints:

• Classification report
• Precision, recall, F1-score
• Overall accuracy

🔹 Comparison Plot

A seaborn boxplot displays CV accuracy distribution across all models.

---

📊 Visual Outputs

The notebook generates:

• Image visualizations (single + multiple)
• Class distribution plot
• Cross-validation accuracy comparison boxplot

---

🎯 Goal of the Project

To evaluate and compare traditional machine learning approaches for MNIST classification without using deep learning, demonstrating:

• Strong baseline model performance
• Benefits of ensemble learning
• Practical ML workflow on image datasets

---

🧩 Technologies Used

• Python 3
• NumPy & Pandas
• Matplotlib & Seaborn
• Scikit-Learn
• XGBoost (imported but not used)

---

🚀 How to Run the Notebook

1. Install required libraries:

   pip install numpy pandas seaborn matplotlib scikit-learn xgboost

2. Run the notebook cell-by-cell.

3. Ensure internet access is available (required to fetch MNIST from OpenML).

---

✨ Author

This project demonstrates classic machine learning techniques applied to MNIST digit recognition.

---

If you want, I can also generate:

• A combined README for both projects
• A GitHub-ready version with badges & images
• A PDF version of the documentation