🍷 Wine Quality Predictor

📜 Project Overview

This project predicts wine quality using Machine Learning techniques based on chemical properties. It explores two primary approaches:

• Regression: Predicts wine quality score (0-10).
• Classification: Categorizes wine quality as low, medium, or high.

Dataset

The Wine Quality Dataset from Kaggle was used. Access it here.

---

🛠️ Features

The dataset includes chemical properties like:

• Alcohol percentage
• pH
• Acidity levels
• Sulfur dioxide content

Data Preparation

• No missing values.
• Balanced dataset by adding samples for underrepresented classes.
• Normalized skewed distributions with log transformations.

Feature Engineering

• Derived Metrics:
  • Alcohol-Sugar Interaction: alcohol × residual sugar
  • Total Acidity: fixed acidity + volatile acidity + citric acid
  • Total Sulfur: free sulfur dioxide + total sulfur dioxide

---

📊 Methodology

Model Building

Random Forest (Classification)

• Handles complex relationships and reduces overfitting.
• Hyperparameters: Tuned using GridSearchCV.
  • max_depth=10
  • min_samples_split=2
  • min_samples_leaf=1

XGBoost (Regression)

• Optimized for large datasets with gradient-boosted trees.
• Hyperparameters: Tuned using Optuna.
  • subsample=0.7449
  • learning_rate=0.0309
  • max_depth=11

---

🚀 Results

• Classification (Random Forest):
  • Accuracy: 85.62%
  • F1 Score: 0.70807
• Regression (XGBoost):
  • R² Score: 0.6028
  • Mean Squared Error (MSE): 0.3350

Key predictors included Alcohol and Total Acidity, as identified through feature importance analysis.

---

Installation

git clone https://github.com/ahmetbekir22/wine-quality-prediction
cd wine-quality-predictor