Knowledge check

Quiz

Complete

Evaluate your understanding of this unit by completing the Knowledge Check.

You're evaluating two classification models for predicting customer churn. Model A has 95% accuracy, but the dataset is highly imbalanced with only 10% churn cases. Model B has 88% accuracy but much higher recall for churned customers. Which model provides greater business value?

Model A, because its accuracy is higher overall.Model A, because high accuracy ensures fewer prediction errors.Model B, because its higher recall means it identifies more actual churners.Model B, because lower accuracy makes it more generalisable.

A healthcare analytics team is building a model to detect early signs of heart disease. What performance metric should they prioritise to minimise the risk of missing true positive cases?

PrecisionRecallAccuracyF1-score

You're optimising a deep learning model that is overfitting. Which strategy most directly improves generalisation without significantly increasing computational cost?

Apply dropout regularisation and early stopping.Add more layers to the model.Increase the batch size dramatically.Train for more epochs to stabilise loss.

An e-commerce company trains a recommendation system that converges very slowly. Which optimisation adjustment is most likely to improve convergence speed without reducing accuracy?

Lower the learning rate significantly.Reduce the training data to make it faster.Remove regularisation to speed up convergence.Increase the learning rate moderately and use learning rate decay.

A data science team compares two models: one performs better on precision, the other on recall. How should they decide which model to deploy?

Choose the model with higher recall, regardless of business needs.Choose the model with higher precision, since it reduces false positives.Analyse which type of error—false positives or false negatives—has greater business impact, and choose accordingly.Average the two metrics and pick the higher overall score.

During model testing, you discover that optimising for F1-score results in slightly lower accuracy but much higher recall. What is the best decision for a fraud detection project?

Revert to the model with the highest accuracy.Keep the F1-optimised model, since it balances precision and recall.Reduce the dataset to simplify training.Increase training epochs to raise accuracy again.

You are designing a new ML workflow for predictive maintenance in manufacturing. Which strategy best balances model accuracy with computational efficiency?

Train on the full dataset every iteration to maintain accuracy.Simplify the model by removing key features.Tune hyperparameters randomly without evaluation.Use cross-validation with early stopping to test performance on multiple subsets efficiently.

A financial institution must ensure its ML models remain efficient and scalable as data grows. Which optimisation approach best supports long-term deployment success?

Focus only on improving current training accuracy.Use manual tuning for each model update.Implement iterative retraining and performance monitoring pipelines.Increase model complexity to capture new patterns automatically.

Your model's compute cost has doubled after hyperparameter tuning, with minimal performance gain. What should you recommend to your team?

Revert to a simpler configuration that delivers near-identical results at lower cost.Continue using the tuned version—it's more advanced.Expand compute resources to handle higher costs.Add more hyperparameters for further tuning.

You're preparing to deploy a model that performs well in testing but consumes significant compute power. The company has sustainability goals. What is the most responsible next step?

Deploy the model immediately since performance is high.Increase training epochs for even better accuracy.Conduct inference optimisation and compare CO₂ emission metrics before deployment.Switch to a smaller model without testing performance trade-offs.

Action item: Knowledge check

In progress