Advanced Data Science and Machine Learning Training Course

Course overview

Many data science teams can build a model that performs well in a notebook but struggle to demonstrate that it will make reliable, commercially useful decisions in production. Advanced practitioners need to choose appropriate modelling approaches, control data leakage, tune models efficiently, explain predictions to stakeholders, and establish monitoring that detects performance decay after deployment. This course addresses the gap between exploratory analysis and defensible machine learning practice, helping participants turn complex data into models that can be evaluated, governed and operationalised.

Participants work through an end-to-end advanced machine learning workflow using Python, Jupyter notebooks and scikit-learn. Topics include feature engineering pipelines, cross-validation design, hyperparameter optimisation, ensemble methods, class imbalance, probability calibration, model explainability with SHAP, experiment tracking, and production monitoring. They learn to compare models against meaningful business baselines, select metrics aligned to decision costs, diagnose overfitting and data drift, and communicate model limitations as clearly as model performance.

Delivery combines instructor-led technical demonstrations with guided coding labs, peer review and a realistic business case. Each participant builds and documents a machine learning solution from raw data through validated model recommendation, including reproducible code, an experiment record, model evaluation results, explainability outputs and a deployment-monitoring plan. This tangible portfolio artefact gives both the participant and their manager evidence of a repeatable approach that can be applied to live analytics initiatives.

The course is designed for experienced analysts, data scientists and technical leads who already use Python for data work and need stronger capability in robust model development, evaluation and governance.

Course objectives

By the end of this course, participants will be able to:

• Design leakage-resistant feature engineering pipelines using scikit-learn Pipeline and ColumnTransformer
• Select validation strategies using stratified cross-validation, time-series splits and nested cross-validation
• Optimise model hyperparameters with RandomizedSearchCV and Bayesian search principles
• Build and compare regularised, tree-based and ensemble models against a business baseline
• Evaluate classification and regression models using cost-sensitive metrics, calibration curves and threshold analysis
• Explain individual and global model predictions using SHAP values and feature importance diagnostics
• Track reproducible model experiments, parameters, metrics and artefacts using MLflow
• Produce a model card and monitoring plan covering drift, performance degradation and retraining triggers

Benefits of attending

Target competencies

Who should attend

• Data Scientists — who need to develop models that are reproducible, explainable and ready for operational review
• Senior Data Analysts — who are moving from descriptive analytics into predictive modelling responsibilities
• Machine Learning Engineers — who need stronger evaluation, experiment-management and monitoring practices
• Analytics Managers — who oversee model delivery and must assess technical quality, risk and business value
• Data Engineering Leads — who design data pipelines that must support reliable feature creation and model retraining
• Quantitative Researchers — who need rigorous validation and model-comparison methods for applied predictive work

Requirements and prerequisites

Participants should be comfortable writing and running Python code, manipulating tabular data with pandas, and creating basic visualisations. They should already understand supervised learning concepts including features, targets, train/test splits, regression, classification, overfitting and common metrics such as accuracy, precision, recall and RMSE. Prior use of Jupyter notebooks and scikit-learn is expected; participants should be able to fit a baseline model independently. Familiarity with Git, SQL, cloud platforms, deep learning or production deployment is helpful but not required. This is not suitable for complete beginners to Python or machine learning.

Training methodology

The five-day programme alternates concise instructor-led technical sessions with guided Python labs in Jupyter notebooks. Participants work on a realistic prediction case, progressively converting raw data into engineered features, validated models, explainability reports and a monitoring design. Demonstrations show the reasoning behind each modelling decision before participants implement it themselves using scikit-learn and MLflow. Structured peer reviews challenge metric choice, validation design and business assumptions. The final workshop converts the completed case into an application plan for a participant's own model, dataset or analytics backlog.

Course outline

Day 1: Robust machine learning problem framing and data preparation

• Translating business decisions into prediction targets and intervention points
• Defining baseline models and decision-relevant success metrics
• Exploratory data analysis for missingness, outliers and target distribution
• Detecting target leakage and temporal leakage in analytical datasets
• Feature typing with numerical, categorical, ordinal and date variables
• Building reproducible preprocessing with Pipeline and ColumnTransformer
• Creating train, validation and holdout datasets for reliable comparison

Workshop: Participants audit a customer-risk dataset, identify leakage risks and produce a reproducible baseline preprocessing and modelling pipeline.

Day 2: Validation design and advanced model development

• Stratified k-fold cross-validation for imbalanced classification
• TimeSeriesSplit and rolling-origin validation for temporal data
• Nested cross-validation for unbiased model selection
• Regularisation with ridge, lasso and elastic net models
• Decision trees, random forests and gradient boosting fundamentals
• Feature interactions and nonlinear transformations
• Bias-variance diagnosis using learning curves and validation curves

Workshop: Participants compare regularised linear and tree-based models using an appropriate cross-validation design and produce a model comparison table.

Day 3: Optimisation, imbalance and decision thresholds

• Hyperparameter search spaces and search-budget design
• RandomizedSearchCV for efficient model optimisation
• Bayesian optimisation principles and sequential search trade-offs
• Class weighting, oversampling and undersampling strategies
• Precision-recall analysis for rare-event prediction
• Probability calibration with Platt scaling and isotonic regression
• Threshold optimisation using confusion-matrix costs and expected value

Workshop: Participants tune a gradient-boosting model, calibrate predicted probabilities and recommend an operating threshold based on a stated cost matrix.

Day 4: Explainability, reproducibility and model governance

• Global feature importance limitations and permutation importance
• SHAP values for global patterns and individual predictions
• Partial dependence plots and individual conditional expectation plots
• Fairness checks across relevant population segments
• MLflow experiment tracking for parameters, metrics and artefacts
• Model cards for intended use, limitations and evaluation evidence
• Communicating uncertainty, error patterns and non-actionable predictions

Workshop: Participants create a SHAP explanation pack and draft a model card that a business sponsor and risk reviewer can assess.

Day 5: Deployment readiness and machine learning monitoring

• Batch scoring and real-time scoring architecture patterns
• Training-serving skew and feature consistency controls
• Data drift, concept drift and prediction drift definitions
• Population stability index and feature-distribution monitoring
• Post-deployment performance measurement with delayed labels
• Retraining triggers, approval gates and rollback criteria
• Model handover checklists and stakeholder decision logs

Workshop: Participants complete an end-to-end model release pack containing evaluation evidence, MLflow records, a model card and a monitoring-and-retraining plan.

Tools & standards covered

Python, Jupyter Notebook, scikit-learn, MLflow

A typical training day

Live online deliveries follow the same structure in the East Africa Time zone, with shorter screen blocks and longer breaks.

What the fee includes

How you can take this course

Certification

Participants who complete the full five days receive the Skillset Development Certificate of Completion, stating the course title, course code, dates and delivery format — suitable for professional-development records and employer reimbursement.

Frequently asked questions