Interpreting Principal Components
What You'll Learn
You'll learn how to read component loadings to understand what each principal component (PC) actually represents in terms of your original features.
The Core Idea
After running PCA, you get new axes (principal components) that capture the most variance in your data. But what are these components? They're weighted combinations of your original features, and the loadings tell you exactly how much each original feature contributes to each PC.
Think of it like a recipe: if PC1 is a smoothie, the loadings tell you how much banana, strawberry, and spinach went into it. A loading of 0.8 for "banana" means PC1 is heavily influenced by that feature, while a loading near 0 means it barely matters.
Reading the Loadings
In scikit-learn, after fitting PCA, the components_ attribute gives you these loadings. Each row is one PC, and each column represents an original feature.
- High positive loading (e.g., 0.7): That feature strongly contributes in the positive direction
- High negative loading (e.g., -0.6): That feature contributes strongly but inversely
- Near-zero loading (e.g., 0.05): That feature barely influences this PC
Practical Interpretation
If you have features [height, weight, age] and PC1 has loadings [0.8, 0.7, 0.1], you might interpret PC1 as capturing "body size" since height and weight dominate. PC2 might capture something else entirely based on its unique loading pattern.
Key Takeaway: Component loadings reveal which original features drive each principal component, turning abstract mathematical transformations into interpretable directions of variation in your data.