Multi-Hazard Disaster Risk Prediction System

A professional grade, real-time machine learning system for assessing flood and heatwave risks using live meteorological data and advanced classification models.

This platform features a premium glassmorphism dashboard, modular backend architecture, and AI-driven risk insights, transforming raw environmental data into actionable disaster intelligence.

---

Problem Statement

Floods and heatwaves are among the most recurring and destructive climate hazards, especially in rapidly urbanizing and climate-sensitive regions. The increasing frequency of extreme weather events demands scalable, data-driven early warning systems.

Traditional forecasting models are often domain-specific and not easily accessible to the public. This project addresses that gap by building a modular, machine learning-based system capable of real-time, multi-hazard risk classification using live environmental indicators.

---

Project Structure

The project has been refactored into a modular architecture to ensure scalability, high performance, and ease of maintenance.

Disaster_Risk_Prediction/
├── app.py                      # Main entry point (Streamlit)
├── requirements.txt            # System dependencies
├── README.md                   # Documentation
├── services/                   # Core Logic & API Integrations
│   ├── api.py                  # OpenWeather & Open-Meteo API handlers
│   ├── model_inference.py      # Model loading and prediction logic
│   └── report_generator.py     # PDF report generation (fpdf2)
├── ui/                         # User Interface Layer
│   ├── theme.py                # Custom CSS & Glassmorphism styles
│   ├── components.py           # Reusable UI components (Gauges, Maps, AI Insight)
│   └── pages/                  # Page-specific rendering logic
│       ├── flood_page.py       # Hydrological risk dashboard
│       ├── heatwave_page.py    # Atmospheric risk dashboard
│       └── about_page.py       # Technical methodology & about
├── utils/                      # Helper Utilities
│   └── helpers.py              # Math, physical risk metrics, & formatting
├── models/                     # Serialized ML Models (.pkl)
├── data/                       # Datasets for training/validation
└── assets/                     # Visual assets & documentation images

---

Key Features

• Multi-Hazard Monitoring: Unified risk assessment for Floods and Heatwaves.
• Hybrid Risk Models: Combines trained ML classifiers (XGBoost/Random Forest) with real-world physical indicators (Heat Index, Elevation).
• Comparison Mode: Side-by-side analysis of two different cities for rapid decision making.
• AI-Powered Insights: Context-aware reasoning for each assessment, explaining risk drivers.
• Professional PDF Reports: Instant generation of downloadable risk reports with metrics and insights.
• Dynamic Charting: Interactive trend analysis using Altair and Plotly gauges.
• Geospatial Intelligence: Real-time map rendering using Folium.

---

Methodology

Flood Risk Module

• Model: XGBoost multi-class classifier.
• Inputs: Rainfall, River Discharge, Water Level, Elevation, Historical Presence.
• Logic: Hybrid scoring using model probability weighted by topographical vulnerability.

Heatwave Risk Module

• Model: Random Forest classifier.
• Inputs: Temperature, Humidity, Wind Speed, Pressure, Rainfall.
• Logic: Atmospheric stress calculation (Heat Index) combined with meteorological trends.

---

Model Performance

Flood Risk Classifier (XGBoost)

Metric	Value
Test Accuracy	98.85%
Macro F1 Score	98.49%

Heatwave Risk Classifier (RF)

Metric	Value
Test Accuracy	96.61%
ROC-AUC	99.23%

---

Deployment Architecture

The system is deployed using Streamlit with a modular front-end/back-end decoupling. The UI is powered by a custom CSS-driven Glassmorphism Theme for a professional control-center aesthetic.

Pages

1. Flood Risk Dashboard

   • Manual city-based assessment.
   • Comparison mode for multi-site monitoring.
   • Hydrological KPI strips and AI insight panels.
   • Interactive Risk Map and PDF report utility.

   

2. Heatwave Risk Dashboard

   • Real-time atmospheric stress analysis.
   • 24-hour temperature projection charts (Altair).
   • Atmospheric risk gauges and AI narrative.

   

3. About & Methodology

   • Detailed breakdown of ML parameters and validation metrics.
   • System architecture layout.

---

Technology Stack

• Framework: Streamlit (Front-end)
• Styling: Custom CSS (Glassmorphism & Dark/Light Modes)
• Machine Learning: Scikit-Learn, XGBoost, Joblib
• Data & Charts: Pandas, NumPy, Plotly, Altair
• mapping: Folium, Streamlit-Folium
• Reporting: FPDF2 (PDF generation)
• APIs: OpenWeather (Weather & Geocoding), Open-Meteo (Elevation & Forecasts)

---

---

How to Run Locally

Prerequisites

• Python 3.9+
• OpenWeather API Key (Free tier)
• Active internet connection

Installation Steps

1. Clone the repository

   git clone https://github.com/logitechsoumili/Disaster_Risk_Prediction.git
   cd Disaster_Risk_Prediction

2. Install dependencies

   pip install -r requirements.txt

3. Configure API Keys The system requires an OpenWeather API key. Create a .streamlit/secrets.toml file in the root directory:

   OPENWEATHER_API_KEY = "your_api_key_here"

4. Launch the Dashboard

   streamlit run app.py

---

Future Enhancements

• Multi-Hazard Expansion: Adding modules for Cyclone, Earthquake, and Drought monitoring.
• Deep Learning Integration: Implementing LSTM/Transformers for temporal flood forecasting.
• Global Coverage: Enhancing geocoding services for broader regional support.
• Automated Alerts: Email and Push notification triggers for high-risk detections.
• Data Persistence: SQL-based historical risk logging and analysis.
• Mobile responsiveness: Optimizing glassmorphism UI for mobile viewing.