Lesson 1Designing target variables: default definitions, delinquency thresholds, time horizonsLearn how to define target variables for credit risk models, including default events, delinquency thresholds, and time horizons, and understand how these design choices affect model performance, calibration, and business relevance in local lending.
Defining default and charge-off eventsChoosing delinquency and DPD thresholdsObservation and prediction window designHandling cures, roll rates, and recoveriesAligning targets with business objectivesLesson 2Customer features to collect: payment history, days past due, invoice-level data, order frequency, average invoice size, sector, size, locationIdentify key customer features for credit scoring, including payment history, days past due, invoice-level behaviour, order frequency, ticket size, sector, size, and location, and learn how to engineer stable, predictive variables from raw data in Zimbabwe.
Payment history and delinquency metricsDays past due and ageing bucketsInvoice-level and transaction featuresOrder frequency and seasonality signalsSector, size, and geographic attributesLesson 3Model interpretability and explainability methods (SHAP, feature importance, monotonic constraints)Learn interpretability techniques for credit risk models, including global and local feature importance, SHAP values, partial dependence, and monotonic constraints, to support regulatory review, customer explanations, and internal trust in Zimbabwean finance.
Global versus local explanationsFeature importance and permutation testsSHAP values for local explanationsPartial dependence and ICE plotsMonotonic constraints in credit modelsLesson 4Augmented features: external data (credit bureau, public filings), macro indicators, trade supplier behaviourExplore how to enrich internal records with bureau data, public filings, macro indicators, and trade supplier behaviour to build robust credit features that improve model accuracy, stability, and early warning capabilities across cycles in Zimbabwe.
Using credit bureau scores and tradelinesIncorporating public filings and legal eventsMacroeconomic indicators for credit cyclesTrade credit and supplier payment behaviourData quality checks and reconciliationLesson 5Bias and risk mitigation: label bias, sample selection, proxy variables, data imbalance and mitigation strategies (regularisation, fairness checks)Examine sources of bias in credit data, such as label bias, sample selection, proxy variables, and class imbalance, and learn mitigation strategies including reweighting, regularisation, fairness checks, and careful feature and policy design for fair lending.
Label bias and historical decision effectsSample selection and survivorship biasProxy variables and sensitive attributesClass imbalance and resampling methodsFairness metrics and monitoring checksLesson 6Common predictive models for credit risk: logistic regression, random forest, gradient boosting (LightGBM/XGBoost)Review common predictive models for credit risk, including logistic regression, random forests, and gradient boosting methods such as XGBoost and LightGBM, comparing strengths, limitations, tuning practices, and deployment considerations in Zimbabwe.
Logistic regression for PD estimationTree-based models and random forestsGradient boosting, XGBoost, LightGBMHyperparameter tuning and validationModel comparison and champion–challengerLesson 7Model governance: validation, backtesting, calibration, population stability, and performance monitoringUnderstand model governance for credit risk, including independent validation, backtesting, calibration, population stability, and ongoing performance monitoring, to satisfy regulatory expectations and maintain reliable production models in Zimbabwe.
Independent model validation scopeBacktesting and benchmarking approachesCalibration and probability of default testsPopulation stability and drift metricsOngoing performance dashboards and alertsLesson 8Decisioning: converting model outputs to risk tiers, credit limits, payment terms and approval workflowsDiscover how to convert model outputs into operational credit decisions, including risk tiers, credit limits, payment terms, and approval workflows, while balancing risk appetite, customer experience, and portfolio profitability in local markets.
Score-to-risk tier mapping and cutoffsSetting credit limits from PD and LGDDesigning payment terms by risk bandApproval workflows and escalation rulesChampion–challenger decision strategies
