Malaysia Flood Risk Prediction System: A Research Project on Interpretability Combining Geospatial Data and Machine Learning

This project focuses on flood risk prediction and interpretability research in Malaysia, integrating geospatial data, rainfall information, hydrological features, and machine learning technologies to provide transparent risk assessment services via a FastAPI backend and Streamlit interactive interface. Key highlights include: interpretable risk scores (displaying influencing factors and weights), Malaysia regional adaptation (compatible with official data sources), and a multi-stage development roadmap, aiming to provide scientific and easy-to-understand flood risk solutions for the public and decision-makers.

Malaysia is located in a tropical monsoon climate zone, where regional flood disasters often occur during the northeast monsoon season from November to March each year. Traditional flood early warning systems rely on the meteorological department's experience judgment and fixed thresholds, making it difficult to accurately assess risks at specific geographic locations. This project aims to build an intelligent risk platform for the entire Malaysia, providing interpretable and quantifiable prediction services through the combination of multi-source data and machine learning. Its core value lies in transparency—not only providing risk scores but also clearly showing risk factors and their weights, helping users understand the source of risks, enhancing public disaster prevention awareness, and assisting emergency decision-making.

Technical Architecture

• Backend: FastAPI (high-performance asynchronous web framework), Pydantic (data validation), GeoPandas+Rasterio (geospatial data processing)
• Frontend: Streamlit (interactive demonstration interface, supporting map visualization)
• Machine Learning: scikit-learn (basic algorithms), XGBoost/LightGBM (prediction models), SHAP (interpretability analysis)
• Data Engineering: Pandas (structured data), MLflow (experiment tracking), DVC (data version control)

Core Functions

1. Transparent Risk Score: Outputs a 0-100 score and risk level (low/medium/high/extremely high), calculated based on multi-dimensional features such as terrain, hydrology, meteorology, environment, and early warnings.
2. Malaysia Adaptation: Coordinate verification ensures input points are within the country, preset examples of major cities, compatible with official data sources like MET Malaysia and InfoBanjir.
3. Interpretability Analysis: Returns confidence level, top 3-5 risk factors, and targeted action recommendations.

The project integrates multi-source public data to build the assessment foundation:

• Official Data Sources: MET Malaysia (meteorological forecast and early warning API), InfoBanjir (public flood information platform), Malaysia administrative division boundary data.
• Satellite and Remote Sensing Data: NASA SRTM DEM (elevation model), Copernicus Land Cover (land cover classification).
• Open Geospatial Data: OpenStreetMap (vector data such as roads and rivers), historical flood inventory (model training and validation).

Iterative development in phases:

• Phase 1 (Completed): Core logic of the transparent scoring engine, FastAPI+Streamlit demonstration application, Malaysia coordinate verification, automated testing and CI/CD processes.
• Phase 2 (Planning): Access to real-time data from MET Malaysia and InfoBanjir, cleaning and validation of historical flood inventory, geospatial feature extraction pipeline.
• Phase 3 (Planning): Construction of grid-based dataset, feature engineering and target leakage prevention, spatial-temporal cross-validation, multi-model comparison (logistic regression, random forest, etc.).
• Phase 4 (Planning): Docker containerization deployment, production environment FastAPI service, complete documentation and API instructions.

• Public Education: Through the Streamlit application, users can input their location to understand risk characteristics and enhance disaster prevention awareness.
• Emergency Response: Assists management departments in quickly screening risks, allocating resources and making evacuation decisions during the rainy season.
• Insurance and Finance: Insurance companies use risk scores to optimize policy pricing and assessment.
• Urban Planning: Provides historical risk data for developers and planning departments to avoid site selection in high-risk areas.

Reference value of this project for similar applications:

1. Interpretability First: In risk-sensitive fields, model interpretability is as important as accuracy.
2. Domain Adaptation: General models need localized adjustments based on specific regional data characteristics and business rules.
3. Data-Driven Iteration: Start with rule-based transparent scoring, then gradually introduce machine learning to improve accuracy.
4. Open Source Collaboration: Complete technical documentation and automated testing lay the foundation for community contributions.

This project is a beneficial attempt of AI in the field of public safety, combining geospatial analysis, meteorological data, and machine learning to provide scientific and easy-to-understand flood risk assessment solutions. With the integration of real-time data and the completion of ML model training, it is expected to become a strong supplement to Malaysia's flood early warning system. For developers and researchers interested in AI for Social Good, disaster risk management, or geospatial intelligence, this is an open-source project worth paying attention to and participating in.