Course Objective & Scope
Many models that perform well in a notebook never reach production because of data leakage, imbalanced classes or decisions nobody can explain. This programme takes teams that already know basic modelling to the level of producing reliable, explainable and traceable models.
Advanced feature engineering, gradient boosting methods, hyperparameter search with Optuna, techniques for imbalanced data, time series forecasting, model explanation with SHAP, and experiment and model registry with MLflow are covered hands-on. Participants model one of the organisation's prediction problems end to end and produce a comparative experiment report and a draft model card.
Who Should Attend?
- Data scientists actively building models
- Machine learning engineers responsible for model quality and release
- Senior analysts providing technical leadership to analytics teams
- Prerequisite: experience with Python, pandas, scikit-learn and basic statistics
Key Takeaways
- Design a cross-validation strategy that prevents data leakage
- Optimise XGBoost/LightGBM models with Optuna
- Choose suitable metrics and resampling techniques for imbalanced classes
- Run backtesting for time series forecasts
- Explain model decisions to business units with SHAP values
- Track experiments, parameters and model versions in MLflow
Curriculum
Advanced Feature Engineering and Validation Strategies
- Target encoding, interaction features
- Group- and time-based cross-validation, data leakage
Gradient Boosting and Hyperparameter Optimisation
- Comparing XGBoost, LightGBM and CatBoost
- Bayesian search with Optuna
Imbalanced Data and Cost-Sensitive Modelling
- PR-AUC, threshold selection and business cost
- Class weights and resampling techniques
Time Series Forecasting
- Trend, seasonality and lag features
- Backtesting and prediction intervals
Explainability and Model Cards
- SHAP and partial dependence
- Documenting assumptions and limits in a model card
Experiment Tracking with MLflow
- Logging experiments, parameters and artifacts
- Model registry and production readiness
Other Courses in This Category
-
Conceptual, logical and physical data modeling, dimensional modeling (Kimball), Data Vault and analytics-oriented data science foundations, covering TDWI body-of-knowledge topics hands-on for data warehouse and BI teams.
- Live Course
- 5 modules
-
Turning business questions into analysis plans, data preparation with SQL and Excel/Python, KPI definition, dashboard design in Power BI or Tableau and data storytelling, for teams moving into or strengthening the analyst role.
- Live Course
- 5 modules
-
Querying and aggregating relational databases with SELECT, WHERE, JOIN, GROUP BY, HAVING and subqueries: hands-on SQL training for business analysts and specialists who want self-service access to raw data.
- Live Course
- 5 modules
-
Window functions, recursive CTEs, cohort and time-series queries, semi-structured (JSON) processing and query tuning with EXPLAIN ANALYZE, for analysts and engineers working on large datasets.
- Live Course
- 5 modules
-
End-to-end data engineering for RAG, LLM and machine learning systems: data contracts, medallion architecture, feature and embedding pipelines, lineage and metadata management.
- Live Course
- 5 modules
-
Automated data quality testing (dbt, Great Expectations, Soda) and enforceable producer-consumer data contracts to stop silent schema changes and corrupted data from breaking downstream analytics and AI.
- Live Course
- 5 modules
Get in touch
