Machine Learning Projects

This repository contains machine learning projects implemented using Python and scikit-learn. The projects cover the implementation of machine learning algorithms, classification tasks, data preprocessing, and model evaluation.

Projects

1. KNN From Scratch

• Implementation of the K-Nearest Neighbors (KNN) algorithm without using machine learning libraries.
• Demonstrates distance calculation, neighbor selection, and majority voting.
• Built to understand the core concepts behind the KNN algorithm.

2. Iris Classification

• Classification using the Iris dataset.
• Train-test split and model evaluation.
• Investigation of the effect of feature scaling on KNN performance.
• Comparison of different distance metrics, including Euclidean, Manhattan, and Cosine distance.

3. MNIST KNN Classification

• Handwritten digit classification using the MNIST dataset.

• Comparison of different values of K.

• Evaluation of multiple distance metrics:

  • Euclidean
  • Manhattan
  • Cosine
  • Chebyshev

• Comparison of search algorithms:

  • Auto
  • Brute Force
  • KD-Tree

• Best Accuracy: 93.37%

Dataset

This project uses the MNIST dataset in CSV format.

Dataset source:

https://github.com/phoebetronic/mnist

Required files:

• mnist_train.csv
• mnist_test.csv

After downloading the dataset, place both files in the project directory before running the notebook.

Note: To reduce computation time, only 10,000 samples from the original 60,000 training images were used in the experiments.

Technologies

• Python
• NumPy
• Pandas
• Matplotlib
• Scikit-learn