Intro to Data Mining: Final Project

Project Overview

This capstone project demonstrates unsupervised learning through clustering analysis of wine varieties. Without labeled data, the goal is to discover natural groupings based on chemical properties.

Problem Statement

Given a dataset of wines with various chemical measurements (acidity, sugar content, pH, alcohol percentage, etc.), use clustering algorithms to identify distinct wine groups and understand what chemical characteristics define each cluster.

Dataset Features

Chemical Properties

• Fixed Acidity: Tartaric acid concentration
• Volatile Acidity: Acetic acid concentration
• Citric Acid: Adds freshness and flavor
• Residual Sugar: Remaining sugar after fermentation
• Chlorides: Salt content
• Free Sulfur Dioxide: Prevents microbial growth
• Total Sulfur Dioxide: Free + bound forms
• Density: Related to sugar and alcohol content
• pH: Acidity/alkalinity measure
• Sulphates: Wine additive
• Alcohol: Percentage by volume

Clustering Approach

Algorithms Implemented

• K-Means Clustering: Partitioning into k clusters
• Hierarchical Clustering: Building a cluster dendrogram
• DBSCAN: Density-based clustering
• Gaussian Mixture Models: Probabilistic clustering

Methodology

1. Data Preprocessing

   • Feature scaling and normalization
   • Dimensionality reduction with PCA
   • Outlier detection and handling

2. Optimal Cluster Selection

   • Elbow method
   • Silhouette analysis
   • Dendrogram visualization

3. Cluster Analysis

   • Profile each cluster by feature means
   • Visualize clusters in 2D/3D space
   • Interpret cluster characteristics

4. Validation

   • Silhouette score
   • Davies-Bouldin index
   • Calinski-Harabasz score

Results & Insights

The project delivers:

• Identification of distinct wine groups
• Chemical profile of each cluster
• Visual representations of clustering results
• Comparison of clustering algorithms
• Recommendations based on cluster characteristics

Key Findings

• Which chemical properties best differentiate wine types
• Natural groupings in the wine dataset
• Relationships between different chemical features
• Quality indicators for different wine clusters

Skills Demonstrated

• Unsupervised learning techniques
• Cluster evaluation and validation
• Dimensionality reduction (PCA, t-SNE)
• Data visualization in high dimensions
• Interpretation of complex results

Real-World Applications

Clustering is used in:

• Customer segmentation for marketing
• Image compression and segmentation
• Anomaly detection
• Recommendation systems
• Document categorization
• Genetic sequence analysis

---

Interactive Notebook