Nigeria's digital lending industry has grown rapidly, expanding access to credit for a population historically underserved by traditional banking. However, this growth has been accompanied by increasing credit risk, while many borrowers remain outside the coverage of traditional credit bureaus.

As a result, digital lenders increasingly rely on alternative behavioral signals—including repayment history, credit utilization, debt burden, income characteristics, and borrower profiles—to assess credit risk.

This project develops a bureau-independent credit risk segmentation framework using the public Give Me Some Credit dataset as a proof of concept.

Although the dataset originates from the United States, its behavioral credit variables closely resemble alternative lending signals commonly used by digital lenders when traditional credit bureau information is limited or unavailable.

The project demonstrates how SQL, Python, and Tableau can be combined to transform raw lending data into actionable business intelligence for:

Borrower default risk segmentation

Financial exposure assessment

Operational risk analysis

Evidence-based underwriting decisions

This project presents an end-to-end credit risk analytics solution covering the complete analytical workflow—from database design and ETL through data quality assessment, feature engineering, SQL analysis, risk scoring, exposure indexing, business interpretation, and executive reporting.

Using behavioral lending data, the project:

Identifies borrowers most likely to default.

Develops a custom Behavioral Risk Score for borrower segmentation.

Builds an Exposure Index to estimate relative exposure where direct loan balance and credit-limit data are unavailable.

Evaluates structural borrower characteristics associated with elevated default risk.

Investigates data-quality anomalies that may contain meaningful predictive signals.

Delivers a four-dashboard interactive Tableau reporting application.

The result is a bureau-independent credit risk framework designed to demonstrate how alternative borrower data can support lending decisions in emerging digital credit markets.

Provides a high-level view of portfolio default risk, behavioral risk tiers, financial exposure, and the compounding relationship between borrower risk and exposure.

Identifies the borrower segments most likely to default and examines behavioral early-warning signals across delinquency history, age, income, and risk tiers.

Evaluates relative borrower exposure using credit utilization, debt burden, and the Exposure Index, while examining whether exposure adds predictive value beyond behavioral risk history.

Examines structural borrower characteristics—including credit line count, income, age, and household dependency—to identify opportunities to strengthen underwriting and improve lending decisions.

Attribute Description Source Kaggle — Give Me Some Credit Original Records 150,000 borrowers Processed Records 149,999 borrowers Dataset Type Consumer Credit Risk Target Variable Serious delinquency within two years (90+ days past due)

Public Nigerian lending datasets containing borrower-level repayment behavior are limited. Therefore, this project uses the Give Me Some Credit dataset as a bureau-independent proof of concept.

The analysis focuses on behavioral lending patterns—including repayment history, credit utilization, debt burden, income characteristics, and borrower profiles—rather than absolute monetary values.

The methodology is therefore designed to demonstrate a transferable analytical framework rather than claim that the source dataset directly represents the Nigerian lending population.

This project answers three core business questions:

Which borrowers are most likely to default, and what are the earliest warning signals?

Where does the portfolio show elevated financial exposure risk, and does exposure add predictive value beyond behavioral history?

Which borrower characteristics and structural patterns indicate opportunities to improve lending decisions?

The project follows a structured analytical workflow:

The workflow combines:

MySQL database design and SQL business analysis

Python ETL and data cleaning

Feature engineering and risk scoring

Behavioral Risk Score development

Exposure Index construction

Exploratory and explanatory visualisation

Interactive Tableau dashboard development

Evidence-based business recommendations

Borrowers aged 20–39 consistently exhibited the highest observed default rates.

A single 30–59 day delinquency was associated with approximately a 5.7× higher observed default rate than borrowers with no such delinquency history.

The Behavioral Risk Score segmented borrowers into Low, Medium, High, and Critical tiers with clear differences in observed default rates.

Placeholder delinquency codes 96/98 produced a 54.65% observed default rate, making them the strongest single risk signal identified in the analysis.

Borrowers in their 30s within the lowest income quartile defaulted at 12.49%, nearly twice the overall portfolio baseline.

A custom Exposure Index was developed to estimate relative exposure using credit utilization and debt burden.

The index is a relative risk proxy rather than a direct measure of monetary exposure because the dataset does not contain loan balance or credit-limit fields.

Two important data-quality limitations were identified and investigated:

The Critical Exposure tier recorded a 22.35% observed default rate, compared with 2.35% for the Low Exposure tier.

Even among borrowers in the Low behavioral risk tier, exposure still separated observed default rates by approximately 6.1×.

Borrowers classified as Critical on both behavioral risk and exposure recorded a 55.0% observed default rate, the highest combined risk segment identified.

Borrowers with zero open credit lines recorded a 25.64% observed default rate, the highest rate among the credit-line bands analysed.

Borrowers with 3–4 dependents in the lowest income quartile defaulted at 15.69%, more than twice the rate of income-matched borrowers with no dependents.

Defaulters consistently differed from non-defaulters across measured characteristics including:

The results support a move away from single-variable approval rules toward multi-factor borrower assessment.

Tool Purpose MySQL Database design, data validation, SQL analysis, and borrower segmentation Python (Pandas & NumPy) ETL, data cleaning, feature engineering, and risk scoring Matplotlib & Seaborn Exploratory and explanatory visualisations Jupyter Notebook Reproducible analytical workflow Tableau Public Interactive business intelligence dashboards Git & GitHub Version control and project documentation

This repository includes:

SQL scripts for database creation, validation, and business analysis

Python notebooks covering ETL, cleaning, feature engineering, and scoring

Processed and feature-engineered datasets

Behavioral Risk Score methodology

Exposure Index methodology

Four-dashboard Tableau reporting application

Dashboard screenshots

Phase-by-phase technical documentation

Data dictionary and project glossary

Portfolio-ready GitHub repository

Comprehensive project documentation is available in the docs/ directory.

Document Description documentation_index.md Documentation overview and navigation phase_00_project_setup.md Project setup, database design, and data ingestion phase_01_data_cleaning_eda.md Data cleaning, quality assessment, and exploratory analysis phase_02_borrower_default_risk.md Borrower default risk analysis and behavioral risk scoring phase_03_financial_exposure_risk.md Exposure analysis and Exposure Index development phase_04_operational_inefficiency.md Operational and structural borrower risk analysis phase_05_dashboard_design.md Tableau dashboard design, architecture, and validation phase_06_business_insights.md Business insights and evidence-based recommendations data_dictionary.md Original dataset fields and engineered variables project_glossary.md Credit risk and fintech terminology security_notes.md Security practices and environment configuration tableau_blueprint.md Dashboard architecture and worksheet mapping

Clone this repository.

Download the Give Me Some Credit dataset from Kaggle.

Place the raw dataset inside data/raw/.

Configure the required database credentials using environment variables.

Execute sql/00_setup.sql to create the database structure.

Run the analytical notebooks in sequence:

Execute the relevant SQL analysis scripts.

Open the Tableau workbook or view the published Tableau Public dashboard.

The project includes an interactive Tableau Public dashboard application consisting of four connected analytical views.

Provides a high-level summary of portfolio risk through:

Portfolio KPI cards

Default risk overview

Exposure tier analysis

Risk × Exposure heatmap

Portfolio population distribution

Executive insights and recommendations

Analyses borrower behaviour and identifies early-warning signals through:

Behavioral risk tier distribution

Default rates by borrower segment

Delinquency data-quality flag analysis

Income and age segmentation

High-risk borrower profiles

Delinquency progression analysis

Evaluates relative exposure risk using:

Exposure Index distribution

Credit utilization analysis

Debt ratio segmentation

Missing-income analysis

Exposure tier performance

Exposure comparison within the Low behavioral risk tier

Identifies structural borrower patterns that may indicate opportunities to improve lending decisions through:

Defaulter versus non-defaulter comparisons

Credit-line analysis

Household dependency analysis

Income segmentation

Operational risk indicators

Interactive filters

Cross-dashboard navigation

Dynamic KPI cards

Custom tooltips

Mark-level annotations

Drill-down analysis

Consistent visual design across all four dashboards

Executive-ready insight and recommendation panels

Tableau Public: View the interactive dashboard

This project is intended for educational and portfolio purposes.

The original Give Me Some Credit dataset is provided by Kaggle under its applicable terms of use.

All SQL scripts, Python notebooks, feature engineering, documentation, dashboards, and analytical methodologies contained in this repository are original work created by Taofeek Salami.