WineWise

Analyze the chemical properties of red wines to identify key factors influencing quality and help winemakers optimize production and improve product appeal.

Table of Contents

• Introduction
• Dataset
• Project Workflow
• Visualizations
• Models Used
• Results
• Future Scope
• How to Use
• Acknowledgments

---

Introduction

WineWise focuses on predicting the quality of red wine samples based on their physicochemical properties. The dataset includes features such as acidity, sugar content, and alcohol levels, and labels wines with a quality score ranging from 0 to 10.

---

Dataset

The dataset contains physicochemical properties and sensory quality ratings of red wine. The key features include:

• Fixed Acidity: Non-volatile acids (g/L)
• Volatile Acidity: Evaporating acids (g/L)
• Citric Acid: Adds freshness (g/L)
• Residual Sugar: Remaining sugar after fermentation (g/L)
• Chlorides: Salt content (g/L)
• Free Sulfur Dioxide: Unbound SO2 (mg/L)
• Total Sulfur Dioxide: Total SO2 (mg/L)
• Density: Mass per unit volume (g/cm³)
• pH: Acidity level
• Sulphates: Prevents spoilage (g/L)
• Alcohol: Percentage by volume (% vol)
• Quality: Quality score (0 to 10)

---

Project Workflow

1. Data Cleaning: Handling duplicates and ensuring consistent data.
2. Feature Analysis: Visualizing correlations between chemical properties and wine quality.
3. Model Training:
   • Decision Tree
   • Random Forest
4. Evaluation: Comparing accuracy scores for both models.

---

Visualizations

• Wine Quality Scores: Most wines have a score of 5, with a slight left skew.
• Correlation Metrics:
  • Alcohol: Moderately positive correlation with quality (0.48).
  • Volatile Acidity: Negative correlation (-0.39).
  • Fixed Acidity: Weak positive correlation (0.12).
  • Free Sulfur Dioxide: Negligible effect (-0.05).
• Residual Sugar vs Quality: Minimal difference between low and high-quality wines.

---

Models Used

1. Decision Tree:
   • Accuracy: 50.7%
   • Pros: Simple and interpretable.
   • Cons: Overfits with complex data.
2. Random Forest:
   • Accuracy: 66.9%
   • Pros: Handles overfitting better and provides robust predictions.
   • Cons: Computationally intensive.

---

Results

The Random Forest model outperformed the Decision Tree model in predicting wine quality, achieving an accuracy of 66.9%, compared to 50.7% for the Decision Tree.

Key insights include:

• Alcohol Content: Strongly correlated with higher wine quality.
• Volatile Acidity: Negatively affects wine quality.
• Residual Sugar: Has minimal impact on wine quality.

---

Future Scope

WineWise has potential for further development in the following areas:

1. Feature Engineering:
   • Investigate non-linear relationships between features and wine quality.
   • Include sensory data like aroma or visual appeal for a holistic prediction.
2. Model Optimization:
   • Fine-tune hyperparameters for Random Forest.
   • Experiment with advanced models like Gradient Boosting or Neural Networks.
3. Data Expansion:
   • Incorporate data from white wines or other wine varieties.
   • Use larger datasets to improve model generalization.
4. Deployment:
   • Create an API or web application to allow winemakers to input wine properties and predict quality in real-time.
5. Explainability:
   • Use techniques like SHAP values to better interpret model predictions for end users.

---