[Day 21] Unsupervised Machine Learning Type 4 - Principal Component Analysis(PCA) (with a Small Python Project)

🎯 What is PCA?

PCA (Principal Component Analysis) is a technique used to reduce the number of features in your data without losing the important information.

Think of it like:

“I have 8 columns of data. Can I shrink it to 2 or 3 powerful ones that explain most of what’s happening?”

That’s exactly what PCA does:

• It compresses your data smartly.
• Keeps the patterns.
• Helps in faster processing, better visualization, and reducing noise.

🌍 Why is PCA Useful in Real Life?

✅ When your dataset has too many columns (a.k.a high-dimensional).

✅ When features are correlated or repetitive.

✅ When you want to visualize complex data in 2D or 3D.

See video below for better understanding:

🔍 Real-World Examples

🏦 1. Fraud Detection (Banking)

• 8+ transaction attributes like amount, device score, time, etc.
• PCA compresses them into 2 key components.
• Helps visualize and flag abnormal behavior.
  

🧬 2. Healthcare - Disease Pattern Recognition

• Thousands of gene expressions for patients.
• PCA helps extract the 2–3 most significant patterns.
• Aids in clustering or diagnosis.
  

🛍️ 3. Customer Behavior Analysis

• Retailers track browsing, purchases, app activity.
• PCA simplifies the customer profile while retaining behavioral signals.
  

🖼️ 4. Image Compression / Face Recognition

• An image has thousands of pixels.
• PCA converts them into a few ‘Eigenfaces’ — compressed faces that still carry identity!

🔧 How PCA Works (Step-by-Step)

Step 1: Standardize the Data

👉 Features must be on the same scale.

Step 2: Find Directions of Maximum Variance

👉 PCA finds the axes (principal components) where data varies the most.

Step 3: Project the Data onto New Axes

👉 The data is re-expressed using fewer features.

🎯 Result: Smaller data, same meaning. Ready for clustering, anomaly detection, or modeling.

🖥 Python

Mini Project: SmartSqueeze: PCA-Based Financial Data Compression

Imagine you're a data analyst at a fintech company. You’ve received transaction logs containing 8 features per transaction like:

• Amount
• Hour of day
• Device Score
• Whether the user has previously committed fraud
• Risk score of the merchant
• Customer tenure
• Number of transactions in the last 30 days
• Location variance

That’s a lot of dimensions to look at — especially if you want to detect patterns or visualize user behavior.

➡️ So you use PCA to reduce it to just 2 powerful features (PCA1 & PCA2).

Dataset:

📊 Enhanced Transaction Dataset (35 Rows × 8 Features)

You can save it as enhanced_transaction_data.csv

📝 Python Code:

import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.preprocessing import StandardScaler
from sklearn.decomposition import PCA

# Step 1: Load the data
data = pd.read_csv("enhanced_transaction_data.csv")

# Step 2: Select features for PCA
features = ['Amount ($)', 'Hour', 'Device Score', 'Prev Fraud (0/1)',
            'Location Variance', 'Merchant Risk Score',
            'Customer Tenure (Years)', 'Num Transactions (30d)']
X = data[features]

# Step 3: Standardize the features
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)

# Step 4: Apply PCA to reduce to 2 components
pca = PCA(n_components=2)
components = pca.fit_transform(X_scaled)
data['PCA1'] = components[:, 0]
data['PCA2'] = components[:, 1]

# Step 5: Visualize the PCA projection
plt.figure(figsize=(10, 6))
sns.scatterplot(
    x='PCA1',
    y='PCA2',
    data=data,
    hue='Prev Fraud (0/1)',
    palette='coolwarm',
    s=100,
    edgecolor='black'
)
plt.title('PCA Projection of Transaction Data')
plt.xlabel('Principal Component 1')
plt.ylabel('Principal Component 2')
plt.grid(True)
plt.legend(title='Previous Fraud')
plt.tight_layout()
plt.show()

Result:

🔬 What Do PCA1 and PCA2 Represent?

• PCA1 (Principal Component 1): This new feature captures the maximum variation in the data. It’s a combination of the original 8 features.
  
• PCA2 (Principal Component 2): This captures the second most important direction of variation, uncorrelated with PCA1.

Together, PCA1 and PCA2 give you a compressed view of your data — 2 dimensions that explain the most meaningful behavior across all 8 features.

They are not just one feature like "Amount" or "Hour" — they are mathematical combinations of all features weighted by their importance.

🔎 What the PCA Graph Tells You:

• Each point = a transaction
• Position in the PCA1–PCA2 space = summary of all 8 features
• Color = fraud label (0 = No fraud, 1 = Fraud)

✅ Insights You Can Get:

• Fraudulent transactions may appear as separate clusters or outliers.
• Transactions with similar patterns are grouped together.
• You can zoom in on suspicious zones or investigate clusters.

Nutshell

1. PCA1 & PCA2 are not original features, but powerful summaries of your data.
2. PCA helps you visualize high-dimensional data and find meaningful patterns.
3. It’s advantageous before applying clustering (like DBSCAN) or anomaly detection.

💬 Join the DecodeAI WhatsApp Channel for regular AI updates → Click here