COMP0173 AI for Sustainable Development Coursework 2 – POVERTY PREDICTION IN NIGERIA USING GEOSPATIAL AND HOUSEHOLD INDICATORS
# AISD Poverty Prediction – Nigeria Context (UCL MSc AI for Sustainable Development)
This repository contains the code used to generate the results presented in my coursework poster:
**“Poverty Prediction in Nigeria Using Geospatial and Household Indicators”**
## Notebooks
- **01_Baseline_Replication.ipynb**
Reproduces the approach from Jean et al. (2016) using proxy satellite-based features.
- **Nigeria_tabular_model.ipynb**
Implements the adapted XGBoost model using LSMS 2018 household tabular features.
Includes:
- data preparation
- log-transform of consumption
- train/validation/test split
- model training
- R² and RMSE evaluation
- predicted vs actual plot
- **01_env_check.ipynb**
Simple environment test (not required to run models).
## Data Availability
The LSMS 2018 dataset is **not included** in this repository due to licensing restrictions.
Notebooks assume the cleaned dataset is available locally:
```python
cons = pd.read_csv("totcons_final.csv")
hh = pd.read_csv("nga_householdgeovars_y4.csv")
df = cons.merge(hh, on="hhid", how="inner")
```
## ⚙️ Environment
The analysis uses the following Python libraries:
pandas
numpy
scikit-learn
xgboost
matplotlib
Tested in Google Colab
📊 Poster
Outputs from these notebooks (figures,and performance metrics) appear in the accompanying poster for:
Coursework 2 — AI for Sustainable Development, UCL
The poster summarises:
The adaptation of the Jean et al. (2016) methodology
The improved performance of LSMS + XGBoost
SDG alignment
Ethical reflections