ML Standard Deviation

ML Standard Deviation and Variance: Measuring Spread in Data

Standard deviation measures how spread out numbers are around their mean. Two datasets can share the exact same average of 50, yet one might stay tightly clustered (48–52) while the other swings wildly (5–95).

In Machine Learning, standard deviation powers feature scaling (Z-score standardization), outlier detection, and model error diagnostics.

Calculating Variance and Standard Deviation in Python

Variance averages the squared distance from the mean, while standard deviation takes the square root of variance so the metric shares the original units (dollars, milliseconds, kilograms).

import statistics

team_a = [80, 82, 79, 81, 80]
team_b = [50, 95, 65, 100, 90]

print("Team A stdev:", round(statistics.stdev(team_a), 2))   # 1.14 (tight)
print("Team B stdev:", round(statistics.stdev(team_b), 2))   # 21.10 (wide)
print("Team B variance:", round(statistics.variance(team_b), 2))

Sample (stdev) vs Population (pstdev)

Python's statistics.stdev() divides by N - 1 (sample standard deviation, matching Pandas .std()), whereas statistics.pstdev() divides by N (population standard deviation, matching default numpy.std()).

Common Mistakes to Avoid

• Mixing np.std() and statistics.stdev() without checking ddof: Set np.std(data, ddof=1) when you want sample standard deviation matching statistics.stdev().
• Confusing variance units with data units: If values are in seconds, variance is in "seconds squared," which is why engineers interpret standard deviation instead.
• Applying the empirical 68-95-99.7 rule to non-normal data: Standard deviation rules assume a roughly symmetric bell curve; skewed data breaks those percentages.

Frequently Asked Questions

What does a standard deviation of 0 mean?

Every number in the dataset is identical, meaning the feature carries zero variation and gives an ML model no predictive signal.

How do I spot outliers using standard deviation?

Compute a Z-score (x - mean) / stdev. Values with abs(z) > 3 sit more than three standard deviations from the average and warrant inspection.

Why does ML care about standard deviation across columns?

Algorithms like KNN, SVM, and linear models with regularization misbehave when one column spans 0–1 and another spans 0–1,000,000.