← Statistics for Data Science learning pathStatistics for Data Science · Module 10 of 10
Statistics for machine learning
Apply statistical thinking to models.
📖 Key ideas
- Train/test split checks generalisation.
- Bias-variance trade-off: too simple underfits; too complex overfits.
- Cross-validation gives more reliable performance estimates.
- Classification metrics: accuracy, precision, recall, F1.
- Data leakage gives falsely good results.
💡 Real-life example: Overfitting is memorising last year's question paper; good models understand the concepts.
OverfittingModel memorises training data
Cross-validationRotating train/test splits
PrecisionCorrect positives among predicted positives
RecallFound positives among actual positives