Aralin 1Pagdidisenyo ng target variables: default definitions, delinquency thresholds, time horizonsMatuto kung paano i-define ang target variables para sa credit risk models, kabilang ang default events, delinquency thresholds, at time horizons, at unawain kung paano nakakaapekto ang mga design choices na ito sa model performance, calibration, at business relevance.
Pagdefine ng default at charge-off eventsPagpili ng delinquency at DPD thresholdsObservation at prediction window designPaghawak ng cures, roll rates, at recoveriesPag-aayon ng targets sa business objectivesAralin 2Mga customer features na kukuhanin: payment history, days past due, invoice-level data, order frequency, average invoice size, sector, size, locationMatukoy ang key customer features para sa credit scoring, kabilang ang payment history, days past due, invoice-level behavior, order frequency, ticket size, sector, size, at location, at matuto kung paano gumawa ng stable, predictive variables mula sa raw data.
Payment history at delinquency metricsDays past due at aging bucketsInvoice-level at transaction featuresOrder frequency at seasonality signalsSector, size, at geographic attributesAralin 3Model interpretability at explainability methods (SHAP, feature importance, monotonic constraints)Matuto ng interpretability techniques para sa credit risk models, kabilang ang global at local feature importance, SHAP values, partial dependence, at monotonic constraints, upang suportahan ang regulatory review, customer explanations, at internal trust.
Global versus local explanationsFeature importance at permutation testsSHAP values para sa local explanationsPartial dependence at ICE plotsMonotonic constraints sa credit modelsAralin 4Augmented features: external data (credit bureau, public filings), macro indicators, trade supplier behaviorGalugarin kung paano pagyamanin ang internal records gamit ang bureau data, public filings, macro indicators, at trade supplier behavior upang bumuo ng robust credit features na nagpapabuti sa model accuracy, stability, at early warning capabilities sa buong cycles.
Paggamit ng credit bureau scores at tradelinesPagdagdag ng public filings at legal eventsMacroeconomic indicators para sa credit cyclesTrade credit at supplier payment behaviorData quality checks at reconciliationAralin 5Bias at risk mitigation: label bias, sample selection, proxy variables, data imbalance at mitigation strategies (regularization, fairness checks)Suriin ang mga pinagmulan ng bias sa credit data, tulad ng label bias, sample selection, proxy variables, at class imbalance, at matuto ng mitigation strategies kabilang ang reweighting, regularization, fairness checks, at maingat na feature at policy design.
Label bias at historical decision effectsSample selection at survivorship biasProxy variables at sensitive attributesClass imbalance at resampling methodsFairness metrics at monitoring checksAralin 6Common predictive models para sa credit risk: logistic regression, random forest, gradient boosting (LightGBM/XGBoost)Suriin ang common predictive models para sa credit risk, kabilang ang logistic regression, random forests, at gradient boosting methods tulad ng XGBoost at LightGBM, na nagkukumpara ng strengths, limitations, tuning practices, at deployment considerations.
Logistic regression para sa PD estimationTree-based models at random forestsGradient boosting, XGBoost, LightGBMHyperparameter tuning at validationModel comparison at champion–challengerAralin 7Model governance: validation, backtesting, calibration, population stability, at performance monitoringUnawain ang model governance para sa credit risk, kabilang ang independent validation, backtesting, calibration, population stability, at ongoing performance monitoring, upang matugunan ang regulatory expectations at mapanatili ang reliable production models.
Independent model validation scopeBacktesting at benchmarking approachesCalibration at probability of default testsPopulation stability at drift metricsOngoing performance dashboards at alertsAralin 8Decisioning: pagkonbert ng model outputs sa risk tiers, credit limits, payment terms at approval workflowsMatuklasan kung paano i-konvert ang model outputs sa operational credit decisions, kabilang ang risk tiers, credit limits, payment terms, at approval workflows, habang nagbabalanse ng risk appetite, customer experience, at portfolio profitability.
Score-to-risk tier mapping at cutoffsPag-set ng credit limits mula sa PD at LGDPagdidisenyo ng payment terms by risk bandApproval workflows at escalation rulesChampion–challenger decision strategies
