https://github.com/rajnandinithopte/machine-learning_knn-classification

KNN classification project for predicting abnormalities in a vertebral column dataset using various distance metrics, EDA, hyperparameter tuning, learning curves, and weighted voting to optimize performance.
https://github.com/rajnandinithopte/machine-learning_knn-classification

Last synced: about 1 month ago
JSON representation

KNN classification project for predicting abnormalities in a vertebral column dataset using various distance metrics, EDA, hyperparameter tuning, learning curves, and weighted voting to optimize performance.

• Host: GitHub
• URL: https://github.com/rajnandinithopte/machine-learning_knn-classification
• Owner: rajnandinithopte
• Created: 2025-02-03T07:40:37.000Z (10 months ago)
• Default Branch: main
• Last Pushed: 2025-02-03T20:59:42.000Z (10 months ago)
• Last Synced: 2025-02-03T21:35:04.134Z (10 months ago)
• Language: Jupyter Notebook
• Homepage:
• Size: 1.23 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Machine Learning: KNN Classification on Vertebral Column Data

## 🔷 Overview

This project applies **K-Nearest Neighbors (KNN) classification** to the **Vertebral Column Data Set**, a biomedical dataset that categorizes spinal conditions into **normal** and **abnormal** classes. The implementation involves **data preprocessing, exploratory analysis, model training, evaluation, and experimentation** with different distance metrics and voting methods.

---

## 🔷 Dataset Description

The **Vertebral Column Data Set** is obtained from the [UCI Machine Learning Repository](https://archive.ics.uci.edu/ml/datasets/Vertebral+Column). It consists of **310 samples** with **6 biomechanical attributes** extracted from radiographic images of the spine.

### 🔶 Features in the Dataset

- Pelvic incidence

- Pelvic tilt

- Lumbar lordosis angle

- Sacral slope

- Pelvic radius

- Grade of spondylolisthesis

Each row represents a **patient's spinal measurements**, and the **target variable** classifies the patient as:

- **Normal (0)**

- **Abnormal (1)** (includes conditions like herniated discs and spondylolisthesis)

---

## 🔷 Libraries Used

To execute this project, the following Python libraries were used:

- **pandas** → Data manipulation and preprocessing

- **numpy** → Numerical computations

- **matplotlib** → Data visualization

- **seaborn** → Statistical data visualization

- **scikit-learn** → Machine learning algorithms (KNN classifier, distance metrics, model evaluation)

## 🔷 Steps Taken to Accomplish the Project

### 🔶 1. Data Preprocessing and Exploratory Data Analysis (EDA)

- Converted categorical class labels into binary labels (Normal=0, Abnormal=1).

- Conducted scatterplot analysis to visualize relationships between independent variables.

- Created boxplots to identify outliers and distribution differences across the two classes.

- Split the dataset into training (first 70 rows of Class 0 and first 140 rows of Class 1) and test sets (remaining data).

### 🔶 2. K-Nearest Neighbors (KNN) Classification

- Implemented KNN with Euclidean distance using either a custom algorithm or scikit-learn’s implementation.

- Evaluated test errors for different values of k within `{208, 205, ..., 7, 4, 1}`.

- Selected the optimal k by plotting training and test errors vs. k.

- Computed performance metrics for the optimal k:

  - Confusion Matrix

  - True Positive Rate (Recall)

  - True Negative Rate

  - Precision

  - F1-Score

### 🔶 3. Learning Curve Analysis

- Investigated the effect of training size on KNN performance.

- Trained the model using different training set sizes `N = {10, 20, 30, …, 210}`.

- Selected the optimal k dynamically for each training size.

- Plotted the learning curve (test error vs. training set size) to analyze model generalization.

### 🔶 4. Experimentation with Different Distance Metrics

- Replaced Euclidean distance with alternative distance measures:

  - Minkowski Distance `(p = 1 → Manhattan, log₁₀(p) ∈ {0.1, 0.2, …, 1}, p → ∞ → Chebyshev)`

  - Mahalanobis Distance (accounting for feature correlations)

- Compared test errors across distance metrics and summarized results in a table.

### 🔶 5. Weighted Voting in KNN

- Implemented distance-weighted voting, where closer neighbors contribute more to the decision.

- Compared performance with Euclidean, Manhattan, and Chebyshev distances.

- Identified the best test error with `k ∈ {1, 6, 11, …, 196}`.

### 🔶 6. Final Evaluation

- Reported lowest training error achieved across all experiments.

- Summarized findings on the best k-value, distance metric, and voting method.

  ---

## 📌 **Note**

This repository contains a **Jupyter Notebook** detailing each step, along with **results and visualizations**.