Practical Work, Project Development and Case Studies
What this domain tests
Practical Work, Project Development and Case Studies tests whether you can run a small end-to-end project on a real dataset: define a problem, clean data, explore it, build a baseline, train and evaluate a model, and communicate results.
Core rules and concepts
- Project lifecycle: problem definition, data collection, cleaning, exploratory analysis, feature engineering, model training, evaluation, deployment or reporting.
- Always create a simple baseline, such as predicting the majority class, so the model's value is measurable.
- Split data into training and test sets before fitting preprocessing. Use validation or cross-validation for tuning. Keep notebooks and scripts reproducible with fixed random seeds and documented versions.
- Choose metrics that match the problem: recall for disease screening, precision when false alarms are costly, F1 when both matter.
Worked example
Predict whether a student passes from attendance and study hours. Clean missing attendance with the median. Use a majority-class baseline, then logistic regression. On the test set, suppose the confusion matrix gives TP = 18, FN = 4, FP = 3, TN = 25. Recall = 18/22 = 0.818; precision = 18/21 = 0.857. Reasoning: recall matters if missing a likely passer needs support; precision matters if support resources are limited.
Traps that cost marks
Data leakage, such as scaling before splitting or using the target as a feature, inflates results. Reporting only accuracy hides class imbalance. Skipping the baseline makes a weak model look useful. Tuning on the test set is a leakage error. A project without documented data source and preprocessing cannot be reproduced.