Wine Quality Prediction

Machine learning analysis of wine quality prediction using physicochemical properties. Complete end-to-end pipeline from data exploration to model optimization.

Overview

This project analyzes the UCI Wine Quality dataset to predict wine quality scores (3-8) based on 11 physicochemical features. The analysis compares Linear Regression and XGBoost models, demonstrating significant performance improvements through feature engineering and hyperparameter optimization.

Key Results

• Best Model: XGBoost Optimized achieves R² = 0.6234 (73% improvement over baseline)
• Important Features: Alcohol content, volatile acidity, sulphates, citric acid
• Methodology: Stratified sampling, feature standardization, RandomizedSearchCV optimization

Dataset

• Source: UCI Wine Quality Dataset (Red Wine)
• Size: 1,599 samples, 11 features
• Target: Wine quality scores (3-8 scale)
• Challenge: Imbalanced dataset (82% samples in quality 5-6)

Analysis Pipeline

1. Exploratory Data Analysis - Feature correlations and distributions
2. Data Preprocessing - Stratified train/test split, feature standardization
3. Model Development - Linear Regression baseline, XGBoost implementation
4. Hyperparameter Optimization - RandomizedSearchCV with cross-validation
5. Performance Evaluation - MSE, MAE, R² metrics comparison

Technical Stack

• Python: Data analysis and machine learning
• Pandas/NumPy: Data manipulation and numerical computing
• Scikit-learn: ML models, preprocessing, model selection
• XGBoost: Gradient boosting implementation
• Matplotlib/Seaborn: Data visualization

Model Performance

Model	Test R²	Improvement
Linear Regression	0.3594	Baseline
XGBoost Basic	0.5891	+63.9%
XGBoost Optimized	0.6234	+73.4%

Repository Structure

├── data/
│   └── winequality-red.csv          # Wine quality dataset
├── notebooks/
│   └── 01_wine_quality_analysis.ipynb # Complete analysis pipeline
├── requirements.txt                  # Python dependencies
└── README.md                        # Project documentation

Key Insights

• Alcohol content is the strongest predictor of wine quality
• XGBoost significantly outperforms linear models on this dataset
• Feature standardization is crucial for optimal performance
• Hyperparameter tuning provides meaningful improvements (5.8% boost)
• Model shows good generalization with minimal overfitting

Skills Demonstrated

• End-to-end machine learning pipeline development
• Feature engineering and data preprocessing
• Model comparison and evaluation
• Hyperparameter optimization techniques
• Data visualization and statistical analysis
• Code documentation and reproducible research

---