Data Science and Predictive Modelling Training Course

Participants can profile source data and document its analytical fitness before modelling.

• Define target variables, prediction horizons and unit-of-analysis boundaries
• Profile records, fields, keys and timestamp coverage before transformation
• Join relational tables using primary keys, foreign keys and grain
• Identify data lineage from source system through analytical dataset
• Assess sampling methods, extract dates and population coverage limitations
• Create a data dictionary with field definitions and business rules

Participants can prepare modelling data using explicit cleaning rules and quality checks.

• Detect missing values, duplicates, invalid ranges and inconsistent categories
• Apply imputation strategies matched to variable type and missingness mechanism
• Standardise dates, currencies, units and categorical labels using pandas
• Treat outliers through capping, transformation, investigation or removal decisions
• Validate business rules with assertions, reconciliation totals and exception reports
• Record cleaning decisions in a reproducible data preparation log

Participants can identify patterns, anomalies and candidate predictors through structured exploratory analysis.

• Summarise central tendency, dispersion, skewness and distribution shape numerically
• Compare target outcomes across segments using grouped statistical summaries
• Examine correlations, multicollinearity and nonlinear relationships between variables
• Plot distributions with histograms, density plots and box plots
• Investigate anomalies through drill-down tables and record-level inspection
• Form testable modelling hypotheses from exploratory patterns and domain knowledge

Participants can create clear visual evidence for data quality, patterns and model results.

• Choose chart types matched to comparison, distribution, relationship and trend
• Build matplotlib and seaborn charts with meaningful scales and labels
• Use faceting and segmentation to reveal subgroup behaviour
• Show uncertainty using confidence intervals, error bars and prediction bands
• Apply colour, annotation and reference lines without distorting interpretation
• Construct decision-focused charts for model performance and business impact

Participants can create, transform and select features that improve predictive model usefulness.

• Encode nominal and ordinal categories using one-hot and ordinal encoding
• Scale numeric variables using standardisation, normalisation and robust scaling
• Create ratios, interaction terms, lags and rolling-window features
• Extract calendar features from dates, timestamps and business schedules
• Reduce skewness using logarithmic, Box-Cox and Yeo-Johnson transformations
• Prevent target leakage through time-aware feature availability checks

Participants can apply statistical reasoning when selecting variables, models and evaluation approaches.

• Probability distributions, sampling variation and confidence intervals interpreted for decisions
• Formulate null hypotheses and alternative hypotheses for analytical claims
• Apply t-tests, chi-square tests and non-parametric comparison methods
• Interpret p-values, effect sizes and statistical power appropriately
• Distinguish correlation, association and causal claims in observational data
• Use train, validation and test sets to estimate generalisation

Participants can build and interpret regression models for continuous business outcomes.

• Fit ordinary least squares regression using scikit-learn and statsmodels
• Interpret coefficients, intercepts and residuals against business assumptions
• Diagnose heteroscedasticity, nonlinearity and influential observations using residual plots
• Measure regression accuracy with MAE, RMSE, MAPE and R-squared
• Add regularisation using Ridge, Lasso and Elastic Net regression
• Generate prediction intervals and communicate forecast uncertainty clearly

Participants can build binary classification models and set decision thresholds for operational use.

• Define positive classes, negative classes and misclassification consequences explicitly
• Fit logistic regression for churn, default and fraud predictions
• Interpret odds ratios, predicted probabilities and classification thresholds
• Construct confusion matrices from predicted labels and actual outcomes
• Calculate precision, recall, specificity, F1-score and balanced accuracy
• Adjust decision thresholds using cost matrices and operational capacity

Participants can compare candidate models using suitable validation designs and acceptance criteria.

• Split datasets using random, stratified, grouped and time-based methods
• Run k-fold cross-validation and repeated cross-validation for stable estimates
• Compare ROC curves, precision-recall curves and area-under-curve measures
• Use calibration curves to assess predicted probability reliability
• Benchmark models against baselines, business rules and naive forecasts
• Document acceptance criteria before selecting a production candidate

Participants can train tree-based models and balance predictive accuracy with interpretability.

• Train decision trees using split criteria, depth limits and pruning
• Interpret tree paths, terminal nodes and segment-level predictions
• Build random forests using bagging and feature subsampling
• Apply gradient boosting with learning rate and estimator tuning
• Compare XGBoost-style boosting concepts with scikit-learn gradient boosting
• Control overfitting through depth, minimum samples and validation results

Participants can identify meaningful segments when no labelled outcome variable exists.

• Standardise variables before distance-based clustering methods are applied
• Run k-means clustering and inspect centroid profiles for segments
• Select cluster counts using elbow plots and silhouette scores
• Apply hierarchical clustering and dendrograms for nested segment structures
• Use principal component analysis to reduce correlated feature sets
• Describe segments through size, behaviour, value and operational implications

Participants can prepare time-series data and create forecasts with time-aware validation.

• Resample timestamped data into daily, weekly and monthly periods
• Identify trend, seasonality, cycles and irregular variation visually
• Create lagged values, rolling averages and seasonal indicator features
• Build baseline forecasts using naive, seasonal naive and moving averages
• Fit autoregressive models and explain differencing for stationary series
• Evaluate forecasts using rolling-origin validation and forecast error metrics

Participants can convert unstructured text into usable features for classification and analysis.

• Clean text through case normalisation, punctuation handling and tokenisation
• Remove stop words and apply stemming or lemmatisation appropriately
• Create bag-of-words and TF-IDF feature matrices for models
• Apply n-grams to capture phrases and local word patterns
• Train text classifiers for routing, sentiment or document categories
• Inspect influential terms and errors to improve text model quality

Participants can assemble repeatable preprocessing and modelling pipelines that reduce implementation errors.

• Build scikit-learn Pipelines linking transformations and estimators together
• Use ColumnTransformer for mixed numeric, categorical and text features
• Separate training transformations from test and scoring transformations
• Tune hyperparameters with GridSearchCV and RandomizedSearchCV
• Track parameters, datasets and results in structured experiment records
• Save trained models and preprocessing objects with joblib files

Participants can explain model behaviour and identify governance concerns before operational release.

• Calculate permutation importance for global predictor ranking
• Use partial dependence plots to inspect feature response patterns
• Interpret SHAP values for individual predictions and model behaviour
• Test performance across protected groups and relevant business segments
• Identify proxy variables, disparate impact and inappropriate target definitions
• Document intended use, exclusions, limitations and human review points

Participants can specify how a validated model will be integrated into a business process.

• Choose batch scoring, real-time scoring or analyst-assisted decision workflows
• Define input schemas, output fields and scoring frequency requirements
• Specify APIs, file interfaces and database hand-offs for integration
• Set decision rules linking predictions to actions and escalation paths
• Plan access controls, retention periods and sensitive-data handling
• Prepare deployment acceptance tests for data and prediction outputs

Participants can define monitoring controls that identify data, performance and business-value deterioration.

• Track input drift using distribution tests and population stability indices
• Monitor prediction distributions, score ranges and decision volumes
• Measure delayed outcome performance against established baseline metrics
• Set alert thresholds, review cadences and model ownership responsibilities
• Investigate concept drift, policy changes and upstream data changes
• Define retraining triggers, approval steps and rollback procedures

Participants can present model findings in terms of decisions, trade-offs and measurable impact.

• Frame analytical questions around decisions, users and measurable outcomes
• Translate model metrics into financial, service and risk consequences
• Present assumptions, limitations and uncertainty without technical jargon
• Use model cards to summarise purpose, data and performance
• Prepare stakeholder questions on fairness, controls and operational feasibility
• Write recommendations with owners, actions, measures and review dates

Participants can build, review and present a complete predictive modelling solution for a defined business problem.

• Frame a business prediction problem with measurable decision criteria
• Prepare source data using documented cleaning and feature engineering rules
• Train and compare baseline and candidate predictive models
• Evaluate performance, calibration, fairness and operational decision thresholds
• Produce reproducible notebooks, model documentation and monitoring requirements
• Present findings, limitations and implementation recommendations to stakeholder reviewers