AI Powered Loan Default Prediction in Banking

Install python 3.8 through anaconda by downloading from anaconda.com and python.org for Windows, macOS, or Linux, follow the installer wizard, and verify the version by typing python in your terminal.

Install Apache NiFi on Windows by installing Java, downloading and extracting NiFi, and running NiFi with the nifi.bat script; access the UI at localhost:8443 for drag-and-drop data flows.

Understand data warehouse architecture for banking models, from data sources and staging to data warehouse and data mart, then power bi reporting and ml deployment with etl, cdc, and governance.

Investigate source systems and key tables, including dim customer, dim account, dim branch, and dim loan, to inform AI powered loan default risk modeling with credit scores.

Explore the one-to-many relationships among dim customer, dim account, dim loan, and dim collateral, illustrating how customers hold multiple accounts, loans, and collateral in a hierarchical financial data structure.

Explore datamart tables that aggregate customer metrics for risk assessment and lending decisions, enabling segmentation and insights; and use the loan default time series for forecasting and risk management.

Create the DW and datamart schemas in MySQL Workbench, populate them with tables via predefined SQL scripts, and verify readiness for data analysis and reporting.

Loads csv data into a MySQL data warehouse using Apache NiFi, building data flows for dim and fact tables, performing truncation, csv to json to sql conversion, and validation.

Explore how customer accounts, loans, and collaterals relate, and learn to aggregate this data in the data warehouse and load it into the final data mart for loan default analysis.

Configure and execute a data mart loading pipeline with Apache NiFi, using process groups, a PL/SQL step, and logging to load data into MySQL data warehouse and data mart schemas.

Explore the machine learning model process for loan default prediction, from problem definition and data understanding to preprocessing, feature engineering, training with XGBoost, evaluation, tuning, and deployment in banking.

Recall the loan default problem and import key libraries, load the Datamart dataset, perform exploratory data analysis, handle missing values, encode categoricals, scale features, and prepare train-test splits.

Explore how XGBoost uses gradient boosting of decision trees to sequentially correct residuals, optimize loss, and deliver fast, accurate regression and classification for structured data.

Train a binary loan default model with XGBoost, optimize hyperparameters via GridSearchCV, and evaluate using AUC, confusion matrix, and metrics such as accuracy, precision, and recall.

Explore artificial neural networks, including neurons, layers, forward and backpropagation, and how weights, biases, and activation functions like ReLU and sigmoid drive learning from structured banking data for prediction.

Apply an artificial neural network to predict loan defaults, using an 80/20 train-test split, standard scaling, a two-hidden-layer ann with sigmoid output, and binary cross-entropy optimization.

Explore time series forecasting for loan default prediction using the prophet model, which automatically handles seasonality, missing data, and supports trend and holiday components.

Forecast future loan defaults with time series Prophet, ARIMA, and SARIMA, modeling trends, seasonality, and external regressors to assess risk and guide capital allocation.

Pack and load the trained XGBoost and Keras models using joblib, pickle, and Keras model save/load methods, and validate with test data.