libock-serge/afriland-credit-risk-dashboard

FICO-style credit risk Streamlit dashboard for an Afriland-shaped retail and SME bank in Cameroon. JHU and VGS Applied AI in Finance, Lesson 6. Synthetic data only. # Afriland First Bank — Credit Risk Dashboard (FICO-style, Streamlit) > Submission artefact for **JHU/VGS Applied AI in Finance — Lesson 6, > Assignment 6 (Tier B)**. Author: Serge Libock. > > Repository: github.com A Streamlit fork of Jim Liew's Lesson 6 starter (`fico_dashboard_starter.py`), customized for an **Afriland-shaped retail and SME bank in Cameroon (CEMAC)**. The application loads the full 20,000-row synthetic dataset shipped with the lesson (`data/fico_data_full.csv`, default rate 11.3%), trains three credit-risk models, and lets an underwriter score a borrower interactively through five sliders re-labelled for Afriland data sources. **Synthetic data only.** Distributions are FICO-shaped; the Cameroon-context narrative is calibrated against public CEMAC/COBAC framing but the model coefficients are not Afriland actuals. --- ## What's in the box | Path | Purpose | | --- | --- | | `fico_dashboard_afriland.py` | The customized Streamlit app — the deliverable. Loads the 20K CSV, fits LR/RF/XGB, renders sliders, ROC, feature importance, what-if, and a 5-row FICO-sourcing table. | | `fico_dashboard_starter.py` | Jim Liew's unmodified starter, kept in-repo for transparency / diff. | | `data/fico_data_full.csv` | 20,000-row synthetic borrower dataset (target = `default_24mo`). | | `data/fico_data_sample1000.csv` | 1,000-row sample for quick smoke tests. | | `prompt_log.md` | The Claude-Cowork conversation used to drive the customization. | | `Tier B Summary Memo.docx` / `.pdf` | The accompanying 1–2 page memo. | ## Run it ```bash pip install streamlit scikit-learn xgboost pandas numpy plotly streamlit run fico_dashboard_afriland.py ``` The app launches at . The 20K CSV loads in under a second on a laptop, and the three models fit in roughly three seconds on the first run (cached thereafter via `@st.cache_resource`). ## Five FICO components → five Afriland features | …