Feature Engineering & Model Evaluation

Course outline and learning expectations

This is a tutor-led course. The outline shows what your tutor will cover; teaching materials and examinations are provided directly to enrolled students.

What you will learn

• Explain how feature quality affects machine learning model performance
• Inspect missing-data patterns and apply appropriate imputation methods
• Encode categorical variables, scale numerical variables and transform skewed features
• Create interpretable features from dates, GPS coordinates and operational data
• Build reusable preprocessing pipelines while preventing data leakage
• Interpret confusion matrices, precision, recall, F1-score, ROC curves and AUC
• Apply suitable methods and metrics to imbalanced classification problems
• Improve and document a weak crop-yield model through feature engineering and evaluation

Modules

1. 1

   Feature quality and model performance

   Explain why useful, reliable features can influence model results more than changing algorithms

   Baseline modelsSignal and noiseData leakageFeature quality checks
2. 2

   Missing data and safe preprocessing

   Identify missingness, choose suitable imputation methods and keep evaluation data separate

   Missing-value patternsSimple and group-based imputationMissing indicatorsTrain-test separation
3. 3

   Encoding, scaling and transformations

   Prepare categorical and numerical variables for machine learning models

   One-hot and ordinal encodingStandardisation and normalisationLog and power transformationsColumnTransformer
4. 4

   Creating features from raw data

   Build interpretable features from dates, locations and operational records using African examples

   Date to seasonGPS to districtRatios and countsDomain knowledgeFeature validation
5. 5

   Model evaluation with real stakes

   Interpret confusion matrices and select precision, recall and F1-score according to the cost of errors

   True and false positivesPrecisionRecallF1-scoreDecision thresholds
6. 6

   ROC curves and AUC

   Read ROC curves, calculate AUC and explain what the metric can and cannot show

   Prediction scoresROC curveAUCMetric comparison
7. 7

   Learning from imbalanced data

   Evaluate and improve models for rare events such as disease cases and digital-payment fraud

   Class distributionStratified splitsClass weightsOver-sampling and under-samplingSMOTELeakage-safe pipelines
8. 8

   Capstone: improve a crop-yield model

   Diagnose a weak Course 10 model and improve its performance through better features and evaluation

   Baseline auditFeature planPreprocessing pipelineMetric comparisonResults report