Take-home A — the PCA scaling trap

On the raw breast-cancer measurements, the first principal component appears to explain 98.2% of the variance, and 1 component reaches 95%.

It is an illusion. worst area has a variance around 323,000 while smoothness values sit below 1, so PCA reports the largest unit, not the largest pattern.

After standardizing, the first component explains 44% and 10 components are needed.

PCA knows nothing about units. Features on different scales must be standardized first.