Async review

Instruction and application

In Progress

Recap core topics:

• Unit 3: Bias-Variance Tradeoff Analysis

Unit 3: Bias-variance tradeoff analysis

In Unit 3, you explored…

• Fundamentals of bias & varianceDefine statistical bias as the systematic error from under-fitting, where overly simple models miss key patterns.
• Define statistical variance as error from over-reacting to training-set noise, causing models to generalise poorly.
• Explain the bias–variance trade-off curve, showing how total error splits into bias, variance, and irreducible noise as complexity changes.
• Evaluate bias and variance impacts on model reliability, interpretability, and business alignment.

✨

Mitigation & auditing strategies

• Assess mitigation effectiveness by comparing fairness improvements against any loss in predictive accuracy.
• Implement detection techniques such as cross-validation, regularisation, data augmentation to identify and correct under- and over-fitting.
• Conduct bias–variance–fairness audits on real data to recommend governance-aligned, evidence-based adjustments.

Knowledge check

Let's see how much you can remember about async unit 3.

Try taking the quiz below, remember it's not a test, it's to help you find out if there are any areas you should brush up on.

How would you identify the primary source of error when a model underfits training data?

A) High varianceB) High biasC) Irreducible noiseD) Over-regularisation

Which of the following techniques most typically reduces variance in an overfitting model?

A) Increasing model capacityB) L2 regularisationC) Collecting more featuresD) Simplifying the dataset

Which plot explicitly visualises how total error decomposes into bias, variance, and noise as model complexity changes?

A) Precision–Recall curveB) ROC curveC) Learning curveD) Bias–Variance decomposition curve

Which method helps detect both underfitting and overfitting by averaging performance across data splits?

A) Confusion-matrix breakdownB) Stratified samplingC) k-Fold cross-validationD) Randomised feature selection

Which combination of strategies respectively targets bias reduction and variance control?

A) Dropout; data augmentationB) Increasing model capacity; bagging ensemblesC) Early stopping; boostingD) Feature pruning; learning-rate decay

In progress