Personal Project

Data-Driven Credit Risk Scoring System for SMEs

Upload a transactions CSV, pick engineered signals, and score the SME with interpretable outputs. Everything runs in-browser so you can experiment without backend calls.

Interactive SME Scoring Demo

The widget below mirrors the production flow: pick features, upload ledger data, then generate a PD, risk rating, archetype, and SHAP-style explanation in a single click.

Live Scoring

Score SME

4 features selected

No file chosen. Using a synthetic sample.

Select engineered features to use

Probability of Default

6.3%

Risk rating

C

Archetype assigned

Seasonal Spiker

Rows processed

2,030

SHAP explanation plot

Which engineered signals moved PD
interpretability
Cash conversion cycles
-0.038
Supplier concentration
-0.031
Seasonality strength
-0.021
Late-payment flags
-0.005
SME credit risk scoring

Project Snapshot

  • Built scoring for SMEs using financial ratios, behavioral signals, and PD modeling.
  • Designed features to capture liquidity, leverage, and payment behavior.
  • Produced interpretable outputs suitable for lending decisions.

Overview

A credit risk scoring system aimed at small and medium enterprises, combining structured financials with behavioral data to estimate probability of default and segment borrowers by risk tiers.

Methodology

Feature Engineering

  • Liquidity and solvency ratios (current ratio, interest coverage).
  • Cash flow stability and revenue variability signals.
  • Behavioral/payment history markers.

Modeling & Evaluation

  • Probability of default modeling with holdout validation.
  • Calibration checks and risk tiering for lending cutoffs.
  • Interpretability (feature importance) for underwriting transparency.

Tech & Tools

Python, Pandas, scikit-learn, PD modeling, calibration, reporting.