Wine Customer Segmentation

Kernal PCA & Logistic Regression

A streamlined Machine Learning approach to classify wine varieties based on chemical composition using Dimensionality Reduction and Linear Classification.

---

Technical Specifications

Component	Specification
Dataset	UCI Wine Dataset (178 Samples, 13 Features)
Preprocessing	StandardScaler (Z-score Normalization)
Dim. Reduction	PCA (13D to 2D) retaining max variance
Model	Logistic Regression (Linear Classifier)
Metric	Accuracy Score & Confusion Matrix
Visualization	Linear Decision Boundaries (2D Projection)

---

The Architecture

The project follows a strict pipeline to ensure data integrity and optimal performance:

graph LR
    A[Raw Data] --> B(Standardization)
    B --> C{PCA Transformation}
    C --> D[2 Principal Components]
    D --> E[Logistic Regression Training]
    E --> F[Prediction & Evaluation]

Loading

1. Data Transformation (PCA)

Instead of feeding 13 complex features directly into the model, we use Principal Component Analysis to extract the most informative signals.

• Why? To mitigate the curse of dimensionality and enable 2D visualization.
• Result: Two new features (PC1 & PC2) capture the essence of the dataset.

2. The Classifier (Logistic Regression)

Since the data becomes linearly separable after PCA, Logistic Regression provides a fast, interpretable, and highly accurate solution.

• Outcome: The model draws linear boundaries between the three customer segments.

---

Performance & Results

The combination of PCA and Logistic Regression proved to be highly efficient for this dataset.

Metric	Score	Note
Training Accuracy	~97%	Model fits the data well without overfitting.
Testing Accuracy	100%	Perfect generalization on unseen data.
Dimensionality	2 Features	Massive reduction from original 13 features.

Observation: The visualization demonstrates distinct clusters for Customer Segments 1, 2, and 3, separated clearly by linear decision boundaries.

---