Logistic Regression Comparison: Scratch vs Scikit-Learn

Project Overview

This project implements Logistic Regression in two ways:

1. From scratch using only NumPy and pandas
2. Using Scikit-Learn’s LogisticRegression class

The main goals of this project are not just to compare two implementations, but also to:

• Practice coding of how logistic regression works internally
• Practice structuring a complete data science project from start to finish
• Gain experience in exploring, cleaning, and preparing a dataset

The dataset used is the Wine Quality dataset (binary: red vs white wine) from Kaggle, dataset

Project Structure

├── data/                    
│ ├── data analysis/           # Notebooks for EDA
│ │ └── data_analysis.ipynb
│ ├── processed/               # Processed dataset
│ │ └── clean_wine.csv
│ └── raw/                     # Raw dataset
│   └── winequalityN.csv
│
├── src/                  
│ ├── pycache/               
│ ├── init.py
│ ├── compare_models.py       # Scratch + sklearn models and compare results
│ ├── data_prep.py            # Data preparation (train/test split, shufflw)
│ ├── logisticR_scratch.py    # Logistic Regression from scratch 
│ ├── main_lib.py             # Main script to run sklearn model
│ ├── main_scratch.py         # Main script to run scratch model
│
├── Makefile                  # (Optional) Automate running and instalation
├── report.pdf                # Final report of the project
├── requirements.txt          # Python packages

How to Run

1. Run Logistic Regression from Scratch

python src/main_scratch.py or make scratch

2. Run Logistic Regression with Scikit-Learn

python src/main_lib.py or make library

3. Compare Both Models

python src/compare_models.py or make compare

This script will output metrics like Accuracy, Log Loss, ROC AUC, Confusion Matrix, and generate a ROC Curve plot

Report

The full analysis, results, and conclusions are documented in:

report.pdf