Explore the Ionosphere

Guided project · 14 steps · Principal component analysis

About

Radar returns from the ionosphere (Johns Hopkins). You reduce 34 continuous features to a few components and see whether "good" vs "bad" returns separate in PC space.

Data

Loaded from /datasets/ionosphere/ionosphere.data. No missing values. Labels are for visualization only—PCA itself is unsupervised. Source dataset is licensed under CC BY 4.0 with attribution to UCI.

Task

Standardize features, fit PCA, read a scree plot, project onto PC1–PC2, color by class, and inspect which original attributes load strongly on PC1.

Cell 1

Step 1: Imports

Cell 2

Step 2: Data path

Cell 3

Step 3: Load the CSV

Cell 4

Step 4: Shape and head

Cell 5

Step 5: Class balance

Cell 6

Step 6: Feature matrix and color labels

Cell 7

Step 7: Scale features

Cell 8

Step 8: Fit full PCA (all components)

Cell 9

Step 9: Scree plot

Cell 10

Step 10: Two-component PCA

Cell 11

Step 11: 2D projection scatter

Cell 12

Step 12: Variance captured in 2D

Cell 13

Step 13: PC1 loadings (which features matter)

Cell 14

Step 14: Bar chart of top PC1 loadings