Machine Learning for Credit Risk Python-Beginner to Advanced

Explore data foundations for credit risk by prioritizing data quality and structure before modeling, and build a data dictionary with Python to define targets, predictors, and leakage risks.

Assess data quality and perform cleaning to ensure credit risk models learn from accurate data. Use Python to profile missing values, handle outliers, standardize types and categories, and document governance.

Turn raw credit data into meaningful features that boost risk model performance. Cover static, behavioral, ratio, and time-based features, aggregations, encoding, leakage avoidance, and Python pipelines.

Explore missing data and imputation strategies in credit risk, including types of missingness, simple imputation with mean or median, missing indicators, and Python-based pipelines that avoid leakage.

Explore why categorical variables matter in credit risk and how to encode them in Python pipelines, balancing one-hot, ordinal, and target encoding with governance and stability.

Learn how to split data into training, validation, and test sets, using time-based and stratified approaches to prevent leakage and ensure realistic, regulatory-compliant credit risk models.

Explore how class imbalance in credit risk data affects model performance, evaluation metrics, and governance, and learn resampling and class weighting techniques to improve defaults detection.

Scale and normalize credit risk data to shape model learning. Apply standard, min-max, or robust scaling, guard against outliers and leakage, and consider interpretability and regulatory needs.

Master feature selection and leakage control for credit risk models by identifying true risk signals, managing multicollinearity, prioritizing stability, business meaning, and time-aware decisions.

Build a model-ready credit risk dataset end-to-end by loading, cleaning, feature engineering, handling missing values, and encoding, ensuring leakage-free, time-stable data for regression and classification models.

Learn the probability of default framework and how machine learning ranks risk within the credit decision process, using logistic regression as a baseline.

Logistic regression remains the benchmark for probability of default, offering transparency and interpretability through coefficients and odds ratios, with class weights, regularization, scaling, ROCAUC, KS, precision, recall, and calibration.

Discover how decision trees model probability of default through rule-based risk segmentation, capturing nonlinear relationships with interpretable paths while highlighting overfitting risks and practical complexity controls.

Explore random forests, an ensemble of decision trees trained on random samples and features, to estimate probability of default with improved stability, using bagging, non-linear patterns, and feature importance.

Explore gradient boosting models for probability of default and how sequential trees correct errors to improve performance, while addressing governance, class imbalance, early stopping, and interpretability in credit risk.

Explore CatBoost for PD modelling, a gradient boosting framework that handles categorical data natively with ordered boosting to prevent leakage, reducing preprocessing and improving credit risk performance.

Examine the use of neural networks for probability of default modeling, addressing overfitting, interpretability, data preprocessing, and regulatory governance alongside comparisons with boosting models.

Learn how pd models are evaluated and validated through ranking-based metrics like AUC and KS, calibration, and time-based validation, with governance and benchmarking for credit risk.

Compare and select PD models, not by complexity or fashion, but by consistently balancing discriminatory power, stability, calibration, interpretability, and regulatory acceptance to deliver reliable business value.

Explore loss-given default (LGD) as the economic severity of default and how machine learning captures complex, non-linear recovery dynamics beyond traditional models.

Explore regression-based machine learning for LGD, learn why LGD is a regression problem with bounded, skewed data, and compare tree-based models and CatBoost for robust governance.

Prioritize data preparation for LGD modeling by defining default, selecting the observation point, and setting recovery windows to prevent leakage and bias.

Validate LGD models by assessing accuracy, bias, stability, and economic realism using metrics like MAE and RMSE, ensuring distributional fit and segment-level performance for regulatory readiness.

Learn how IFRS 9 requires point-in-time, forward-looking LGD aligned with PD and EAD, using scenario-weighted forecasts, macro drivers, and governance overlays for recoveries and validation.

Transform LGD models into actionable scores and integrate them into IFRS 9 expected credit loss calculations, ensuring point-in-time, scenario-aware LGD aligned with PD and EAD.

Run an end-to-end LGD workflow for IFRS 9, from portfolio understanding to ECL integration, using ML models built on disciplined data preparation and governance.

Explore exposure at default (EAD) and credit conversion factors (CCF) to model loss exposure, using regression and machine learning in Python for IFRS 9 readiness.

Develop and apply feature engineering techniques for EAD and CCF modeling to capture borrower behavior, utilization, trends, and macroeconomic interactions, while avoiding leakage and ensuring stability.

Explore how exposure at default and credit conversion factor models are evaluated and validated, emphasizing governance, bias, tail risk, segment and time-based testing, and clear explanations for stakeholders.

Integrate EAD and CCF into IFRS 9 expected credit loss calculations, linking stage 1 and stage 2, PD, LGD, and scenario-based EAD to produce consistent, governance-driven ECL outputs.

Build an end-to-end EAD and CCF workflow in Python, from portfolio understanding to IFRS 9 compliant ECL outputs with validation.

Apply the Hosmer Lemeshow test to assess calibration of IFRS 9 probability of default models by comparing observed and expected defaults across deciles.

Assess probabilistic predictions with the Breyer score, a mean squared error between predicted PDs and actual defaults, used for IFRS 9 calibration and credit risk modeling.

Explore the KS statistic for measuring model discrimination in credit risk, detailing calculation steps, interpretation, and its use with cumulative defaults vs non-defaults for IFRS 9 compliance.

Learn to use the ROC curve, AUC, and C-statistic as discrimination metrics for IFRS 9 credit risk models. Interpret thresholds to distinguish defaulters from non-defaulters and compare model performance.

Compare the gini coefficient, lift curve, and gains chart to evaluate a model's discriminatory power in credit risk and IFRS 9 applications.

Rank deciles by sorting accounts by predicted default probability to assess credit risk models, form ten deciles, and evaluate model performance with gains, lift, and IFRS 9PD validation.

Explore model selection criteria in credit risk modeling by comparing negative 2 log likelihood, AIC, and BIC, and understanding their tradeoffs for prediction, interpretability, and IFRS 9 applicability.

Explore the D-Statistic, concordance, and discordance to assess a credit risk model's discrimination. Learn to calculate, interpret, relate them to AUC, and apply IFRS 9 context for validation and monitoring.

Learn how the Population Stability Index (PSI) measures model stability, detects data drift, and flags portfolio shifts in IFRS 9 credit risk models. Learn how PSI is calculated and interpreted.

Validate logistic regression credit risk models with a validation rerun and scoring validation sample to ensure predictive consistency, detect overfitting, and meet IFRS 9 using AUC, KS, Gini, and calibration.

Explore model performance metrics for credit risk under IFRS 9, covering discrimination, calibration, stability, and selection metrics, including auc, gini, ks, d-statistic, concordance, and psi.

Develop a point-in-time IFRS 9 PD model in Python with a 70-30 split to prevent leakage, using logistic regression with preprocessing and evaluating AUC, Gini, KS, brier score, and PSI.

Bridge the structural Merton model with IFRS 9 point-in-time PD requirements to convert market-implied default risk into 12-month and lifetime PIT PDs for listed corporates.

Understand IFRS 9 staging and SICR, and how Basel IRB models adapt with point-in-time PDs, lifetime ECL, and macro overlays for accurate credit risk provisioning.

Learn to evaluate and validate IFRS 9 staging rules across Stage 1 to Stage 3, using KS, GINI, PSI, and transition dynamics to ensure accurate expected credit loss.

Explore a complete IFRS 9 staging engine built in Python and pandas, loading a PIT PD dataset, constructing stage 1–3 rules, and evaluating staging quality.

load the IFRS 9 dataset with pandas, convert it to a data frame, and preview the head to verify file path, structure, and row count before staging and PD analysis.

Detect and standardize IFRS 9 staging columns, convert days-past-due and default flags to numeric, normalize PD12M, and create an auditable preprocessing workflow for reliable modeling.

Clean DPD values, standardize the default flag, normalize PD, prepare the report date, and derive the PD at origination proxy and the PD ratio for IFRS 9.

Explore how IFRS 9 uses PD at origination and the PD ratio to detect SICR, compute migrations to stage 2, and validate a transparent, auditable staging framework.

Evaluate the IFRS 9 staging engine using evaluation A, B, and C to measure default rate by stage, timeliness to default, and lift with PD ranking.

Compute and interpret the population stability index (PSI) to detect shifts between early and latest cohorts' PD distributions, using decile bins for IFRS 9 model monitoring.

Export structured IFRS 9 outputs and documentation to establish auditable, reproducible model governance, enabling traceability and regulator-ready validation in banking risk models.

Integrate the full IFRS 9 staging workflow in a transparent python engine, from data loading and preprocessing to stage 1–3 decisions using DPD, default flags, PD origination, and PD ratio.

Master IFRS 9 staging and evaluation techniques through a macro-based framework that assigns loans to stages and validates model performance with lift tables, transition matrices, and PSI.

Explore lifetime probability of default modeling under IFRS 9 for stage 2 and 3 ECL, covering cohort, survival, and transition matrices with SAS implementations and macroeconomic scenarios.

Learn how lifetime PD drives the calculation of expected credit loss (ECL) by summing monthly PD, LGD, EAD, and discount factors, with an illustrative example.

Compare point-in-time PD and lifetime PD, and examine how IFRS 9 stage 1 to stage 2 transitions create a cliff effect in provisions, shaping risk management and capital planning.

Explore cohort analysis to track groups that start together over time, revealing how defaults and lifetime pd evolve across origination cohorts in credit risk.

Use transition matrices to model how loans move from current to 30 DPD, 60 DPD, or default, and forecast lifetime default under IFRS 9.

Demonstrate an end-to-end lifetime PD and ECL workflow in Python, from a loan-level panel dataset to lifetime ECL estimates, with horizon, lgd, and initial data inspection.

Prepare an analysis-ready loan panel by converting dates, building a monthly time index, and assigning origination cohorts to enable survival curves, lifetime PD, and ECL by cohort.

Transform the raw lifetime data into a clean monthly panel for survival analysis and lifetime PD/ECL modeling by enforcing required columns, converting calendar dates, computing timeM, and assigning origination cohorts.

Convert integer dates to real timestamps, build a stable monthly time index, and prepare the portfolio for survival modeling with Kaplan-Meier time-to-event analysis.

Learn why explainability matters in credit risk, balancing transparency and predictive power for PD, LGD, and EAD, with XAI techniques and regulatory considerations.

Analyze global explainability techniques for credit risk, using SHAP values, feature importance, PDP and ALE to interpret portfolio-level behavior, nonlinear interdependencies, and regulatory validation across PD, LGD, and EAD models.

Explore local explainability in credit risk, distinguishing local from global explanations and supporting audit-friendly, fair decisions with SHAP and LIME.

Explore how explainability varies by model type, from logistic regression to neural networks, and learn governance, regulatory expectations, and post hoc techniques like SHAP across credit risk models.

Learn how regulators and auditors evaluate explainability and demand structured documentation. See how evidence packs with SHAP, PDP, and local explanations support governance and model defense.

Identify and mitigate bias in credit risk models through fairness, ethical ai, and governance. Learn how regulators demand explainable, non-discriminatory decisions and how to monitor fairness across segments.

Explore practical xai with shap in a 12-month pd default model, comparing logistic regression to gradient boosting, generating global and local explanations and translating findings for audit, governance, and regulation.

Build a production-ready training pipeline for credit risk models that is modular, auditable, and automated, with validation, versioning, and reproducible training.

Turn your trained credit risk model into a real-time api service with fastapi, enabling pd, lgd, ead predictions at scale, with input validation, versioning, and secure, documented endpoints.

Integrate machine learning predictions (PD, LGD, EAD) into a production Postgres system of record with input, feature, and output layers, enabling batch and real-time scoring, audit trails, and data lineage.

Develop and deploy a production monitoring dashboard that tracks model stability, feature drift, performance (roc, ks, gini), and realized defaults, with psi-driven alerts and override logging to satisfy regulators.

Master how CI/CD pipelines enable safe, repeatable, auditable updates to ML models in credit risk. Learn about version control, testing, environments, containerization, model registries, governance, and rollback under regulatory frameworks.

Transform raw probability of default data into a professional Power BI dashboard for IFRS 9 risk analytics, with data cleaning, age band categorization, interactive region filters, and published data governance.

Transforms raw credit risk data into Power BI data quality dashboards for Basel 3.1 and FRS 9, highlighting missing values, anomalies, and profiling kpis for reliable pd and expected loss.