Linear Regression Diagnostics Toolkit

This repository implements a modular Python pipeline for analyzing and validating linear regression assumptions using the Body Measurements Dataset. It focuses on statistical rigor through assumption checks, not just predictive performance.

🧠 Overview

The codebase provides tools to test and visualize the four critical assumptions of linear regression:

1. Linearity
2. Independence of errors
3. Homoscedasticity (constant variance)
4. Normality of residuals

All diagnostics are handled with reusable Python modules, no Jupyter notebooks involved.

📂 File Structure

.
├── main.py                    # Entry point for model fitting and diagnostics
├── utils.py                   # Helper functions (e.g., data cleaning, visualization)
├── check_linearity.py         # Linearity test via partial regression plots
├── check_normality.py         # Normality check (Q-Q plot, Shapiro-Wilk test)
├── check_homoscedasticity.py  # Breusch-Pagan & White tests
├── check_independence.py      # Durbin-Watson test and residual autocorrelation
└── data/                      # (Expected) Contains the CSV dataset

📊 Dataset

• Source: Kaggle - Body Measurements Dataset
• Content: Anthropometric data such as age, height, weight, and body part circumferences
• Target Variable: Varies—commonly height, weight, or body fat percentage

---

🚀 How to Run

Make sure you have uv installed:

uv pip install -r pyproject.toml

Then run the diagnostics:

uv run main.py

✨ The pipeline will fit a linear regression model and sequentially check each assumption, printing results and plotting visuals using matplotlib and seaborn.

---

🧪 Diagnostic Modules

Each assumption test is cleanly separated into its own script:

✅ check_linearity.py

• Uses residual plots and added variable plots
• Highlights multicollinearity issues via VIFs

✅ check_independence.py

• Calculates Durbin-Watson statistic
• Plots residuals against time/index order

✅ check_homoscedasticity.py

• Breusch-Pagan and White tests
• Residuals vs fitted plot with confidence bands

✅ check_normality.py

• Shapiro-Wilk test
• Histogram and Q-Q plots

---

📦 Dependencies

All dependencies are defined in pyproject.toml and managed with uv.

Main packages used:

• pandas, numpy
• scikit-learn
• statsmodels
• seaborn, matplotlib
• scipy

---

🎯 Key Highlights

• 🧱 Modular design for each assumption
• 🧪 Automated statistical testing + visual plots
• 🧰 Ready for integration into larger ML workflows
• 🔍 Emphasizes statistical validation before prediction

---