Logo Lanfrica

neneo25/floodguard-nigeria

Domaine:

climateenvironment and energy

Type de record:

model
Créateur:
nen
Hôte:
AI-powered flood risk classification for Nigerian LGAs using rainfall, population and flood-impact data. # FloodGuard Nigeria 🌧️ ## AI-Powered Flood Risk Classifier FloodGuard Nigeria is an AI/ML project that classifies flood-impact risk across selected Nigerian Local Government Areas (LGAs). The project combines historical flood-impact data, rainfall indicators and population data to classify an LGA as **Low, Medium or High risk**. ## Why I Built This Flooding affects many communities across Nigeria, resulting in displacement, loss of livelihoods and other humanitarian impacts. I wanted to explore how machine learning and publicly available data could be used to create a simple tool that helps identify areas with higher flood-impact risk. ## Data Sources The project uses data from: - Humanitarian Data Exchange (HDX) - NEMA flood-impact records - Subnational rainfall data - LGA population data The datasets were cleaned and joined using Nigerian administrative PCODEs. The final modelling dataset contains **40 LGAs and 15 features**. ## Features Used The model uses rainfall and population-related features, including: - Average and maximum 10-day rainfall - Average and maximum 1-month rainfall - Average and maximum 3-month rainfall - Rainfall anomalies - Rainfall variability - 90th percentile rainfall - Extreme rainfall days - Log-transformed population ## Machine Learning I tested three classification models: - Logistic Regression - Random Forest - Decision Tree Logistic Regression produced the best results on the test set and was selected as the final model. | Model | Accuracy | Precision | Recall | F1 Score | |---|---:|---:|---:|---:| | Logistic Regression | 50.0% | 55.4% | 48.9% | 47.8% | | Random Forest | 42.5% | 44.1% | 40.0% | 39.0% | | Decision Tree | 32.5% | 28.3% | 30.0% | 27.5% | Because the modelling dataset is relatively small, these results should be interpreted as a prototype evaluation rather than production-level performance. ## Application The application was built with Streamlit. Open FloodGuard Nigeria A user can: 1. Select a st …