Course Objective & Scope
On tabular data, model success is usually decided by feature quality rather than the algorithm. The programme covers one-hot, ordinal and target encoding for categorical variables, aggregation features over dates and time windows, ratio and interaction terms, feature extraction from text and geographic data, and feature selection with permutation importance and SHAP.
Point-in-time correctness and leakage prevention are stressed throughout. Participants build a reusable feature pipeline on data adapted to the organisation's case and see how to share features through a feature store such as Feast.
Who Should Attend?
- Data scientists and ML engineers
- Data engineers preparing model data
- Risk, marketing and operations analytics teams
- Prerequisite: experience with pandas and scikit-learn
Key Takeaways
- Choose the right encoding for categorical and high-cardinality variables
- Generate time-window aggregation features without leakage
- Measure feature impact with ablation and SHAP
- Build a reusable feature pipeline with scikit-learn ColumnTransformer
- Manage training-serving skew between training and production
Curriculum
Numeric and Categorical Transformations
- Scaling, log and binning transforms
- One-hot, ordinal, target and frequency encoding
Time-Based Features
- Lag, rolling-window and recency-frequency features
- Preventing leakage with point-in-time joins
Interaction, Text and Geographic Features
- Ratio, difference and interaction terms
- Feature extraction from TF-IDF, embeddings and location data
Feature Selection
- Filter, wrapper and embedded methods
- Permutation importance, SHAP and ablation tests
Feature Pipelines and Feature Stores
- Reusable pipelines with ColumnTransformer
- Online/offline feature sharing with Feast and training-serving skew
Other Courses in This Category
-
Choosing the right metrics, evaluation tied to business KPIs, error analysis, fairness checks and model explanation with SHAP, LIME and PDP, so models can be presented to stakeholders reliably and understandably.
- Live Course
- 5 modules
-
Independent model validation: data and conceptual soundness checks, backtesting, stability (PSI), challenger models, stress tests and findings reporting, producing a validation report aligned with a model risk management framework.
- Live Course
- 5 modules
-
Time series forecasting with ARIMA, ETS, Prophet and LightGBM-based approaches, working hands-on with seasonality, exogenous variables, hierarchical series, backtesting and prediction intervals to build reliable forecasts.
- Live Course
- 5 modules
-
Detecting unusual behaviour in unlabelled data with statistical methods, Isolation Forest, LOF, autoencoders and time series anomaly techniques, including threshold setting, alert fatigue and evaluation strategies.
- Live Course
- 5 modules
-
Applied Machine Learning
End-to-end machine learning with Scikit-learn, XGBoost, LightGBM, and CatBoost, covering feature engineering, Optuna hyperparameter tuning, imbalanced data, and SHAP/LIME explainability.
- Live Course
- 3 modules
-
Behavioural features, gradient boosting, anomaly scores and hybrid rule-model setups for payment, insurance and account fraud, with hands-on work on extreme class imbalance, delayed labels and operational threshold management.
- Live Course
- 5 modules
Get in touch
