Statistics for GATE Machine Learning
Statistics asks: what can this sample honestly say about the world? GATE tests estimation, uncertainty and testing — the same ideas behind validation scores and confidence in ML.
Important concepts
- Point estimation, bias, variance, MSE = bias² + variance + noise.
- Maximum likelihood vs MAP; sampling distributions.
- Confidence intervals vs credible statements.
- Hypothesis tests: z, t, chi-square; p-values done honestly.
- Type I/II errors, power, multiple testing.
- Correlation vs causation; Simpson-type reversals.
GATE relevance
Bias–variance decomposition questions, reading confidence intervals, choosing tests, and evaluation questions where accuracy on imbalanced data misleads.
Important formulas
MSE=Bias²+Var+σ² · SE=σ/√n · CI=x̄±z·SE · r=Cov/σxσy
Worked example
Accuracy 90% on 100 points: SE≈√(0.9·0.1/100)=3%, so a 95% interval is roughly 84–96%. Small test sets mean wide uncertainty — the reason GATE rewards cross-validation thinking.
Practice questions
Practice questions (Evaluation unit) + mock test. Theory: Guide Unit 5 and formula sheet.
Related topics
Probability · Linear algebra · Machine learning
FAQ
What is the bias–variance tradeoff?
Simple models underfit (high bias); complex models memorise noise (high variance). Regularisation and validation pick the middle.
When is accuracy misleading?
On imbalanced data a majority-class predictor scores high while learning nothing — use precision, recall, F1 and PR curves.