C01 Stage B · Core AI 56 h (20 T / 36 P)
Classical Machine Learning
Frames, builds and validates supervised and unsupervised models without leakage.
Prerequisite
F01, F03, F04
Modules
6
NOS
—
Statutory mapping
—
Tech stack
scikit-learnXGBoostLightGBMOptuna
Modules
6 modules · 56 hours
M1 · Learning paradigms and framing
8 h TOPICS
- Supervised, unsupervised, semi-supervised, reinforcement
- Problem framing
- Train/validation/test discipline
- Overfitting and underfitting
- Bias–variance decomposition
Lab: Construct leakage-free splits for a temporal dataset
Course material for this module is in production.
M2 · Regression
8 h TOPICS
- Linear and polynomial regression
- Ridge, Lasso, ElasticNet
- Assumptions and diagnostics
- Residual analysis
Lab: Regularisation sweep with learning curves
Course material for this module is in production.
M3 · Classification
10 h TOPICS
- Logistic regression
- k-NN
- Naïve Bayes
- Support vector machines and kernels
- Decision boundaries
- Class imbalance
Lab: Imbalanced classification with resampling and threshold tuning
Course material for this module is in production.
M4 · Trees and ensembles
10 h TOPICS
- Decision trees
- Bagging and random forests
- Gradient boosting, XGBoost, LightGBM
- Stacking
- Feature importance
Lab: Tune a gradient-boosted model against a baseline
Course material for this module is in production.
M5 · Unsupervised learning
10 h TOPICS
- k-means
- Hierarchical clustering
- DBSCAN
- Silhouette analysis
- PCA and dimensionality reduction
- Anomaly detection
Lab: Segment customers and justify the cluster count
Course material for this module is in production.
M6 · Feature engineering and pipelines
10 h TOPICS
- Encoding and scaling
- Binning and interactions
- Temporal features
- Target leakage
- Pipelines and column transformers
- Hyperparameter search
Lab: End-to-end reproducible pipeline object
Course material for this module is in production.
Real-world work scenario
Your model scores 0.94 AUC offline and fails in production. Discover that a feature was computed using post-event information, rebuild the feature set with a strict point-in-time join, and requantify honest performance.
Assessment
Tabular modelling project with evaluation report