MNIST Classification using Machine Learning Models

Project Overview

This project demonstrates the application of machine learning algorithms for classifying the MNIST dataset using a combination of models from scikit-learn and TensorFlow. The MNIST dataset contains images of handwritten digits (0-9), and the goal is to classify these images accurately.

The project includes:

• Data preprocessing using normalization and reshaping.
• Implementation of multiple machine learning models, including:
  • K-Nearest Neighbors
  • Decision Tree Classifier
  • Logistic Regression
  • Random Forest Classifier
• Hyperparameter tuning using GridSearchCV.
• Model evaluation based on accuracy score.

Dataset

The MNIST dataset is loaded directly using TensorFlow's keras.datasets module. The dataset contains 70,000 28x28 pixel grayscale images of handwritten digits. The dataset is split into training and testing sets.

• Training data: 60,000 images
• Test data: 10,000 images

For the purposes of this project, a smaller subset of the data is used with 6,000 images for training and evaluation.

Code Overview

1. Data Loading:

   • The MNIST dataset is loaded using TensorFlow's built-in keras.datasets.mnist loader.

2. Data Preprocessing:

   • The images are reshaped into 1D arrays of size 28x28 = 784 for each image.
   • The features are normalized to improve model performance.

3. Model Training and Evaluation: The following models are trained on the normalized data:

   • K-Nearest Neighbors (KNN)
   • Decision Tree Classifier
   • Logistic Regression
   • Random Forest Classifier

4. Hyperparameter Tuning: GridSearchCV is used to find the best hyperparameters for the K-Nearest Neighbors and Random Forest models, tuning parameters such as:

   • K-Nearest Neighbors: n_neighbors, weights, and algorithm
   • Random Forest: n_estimators, max_features, and class_weight

5. Model Evaluation: The models are evaluated using accuracy metrics, and the best models are displayed along with their tuned parameters.