Scratch Coding Classes for Kids in Heraklion

Chapter 138 – Principal Component Analysis (PCA) Explained

1. What Is PCA?

Principal Component Analysis reduces the number of features while keeping as much data variance as possible. High-dimensional data (e.g. 30 correlated columns) is mapped onto fewer orthogonal axes (principal components) sorted by variance.

2. Four Practical Benefits

• Less overfitting: models learn broad patterns instead of noise
• Faster training & inference: fewer features mean quicker compute
• Noise filtering: low-variance directions are dropped
• Visualization: 10+ dimensions become 2D or 3D plots

3. How Components Are Built

1. PC1 — direction of maximum variance
2. PC2 — next highest variance, orthogonal to PC1
3. Remaining PCs sorted by decreasing variance
4. Keep the top n components (user choice); discard the rest

4. Correlated Features Example

Math and Physics scores often move together. PCA can fold that shared trend into a single component, removing redundancy while retaining the spread of ability across students.

5. Advantages vs Limitations

Type	Point
Advantage	Fewer features, same core variance
Advantage	Faster models and real-time scoring
Advantage	Merges redundant / correlated columns
Advantage	Enables 2D/3D plots of high-D data
Limitation	PCs are linear mixes — harder to interpret
Limitation	Some information is lost with discarded PCs
Limitation	Must standardize features first (e.g. StandardScaler)