DengueRadarSG

Inspiration

Singapore faces recurring dengue outbreaks, but existing public-facing dengue maps mainly show where transmission is already occurring. We were inspired by the possibility of using Singapore's open government data to move from reactive monitoring to earlier risk prediction.

DengueRadarSG aims to answer a simple question:

Which parts of Singapore are most at risk of increased dengue activity over the next two weeks?

By combining dengue activity with rainfall, temperature and geospatial data, we hope to provide earlier, location-specific signals that can help communities and public-health stakeholders prioritise preventive action before case counts escalate.

What it does

DengueRadarSG is a proposed 2-week dengue risk forecasting platform for Singapore.

For each URA planning area, the system will generate a dengue risk score and classify the area as Low, Medium or High risk.

The dashboard will bring together:

Current dengue cluster activity
Recent dengue case trends
Real-time rainfall and temperature readings
Historical weather patterns
Planning-area geography

Users will be able to view an interactive Singapore map showing current clusters, forecasted risk, changes from the previous period and the main factors influencing each prediction.

The goal is not to replace existing dengue surveillance, but to complement it with a forward-looking early-warning layer.

How we will build it

DengueRadarSG is designed as an end-to-end pipeline on Databricks Free Edition.

Data will be sourced from Singapore public datasets, including NEA dengue clusters, MOH historical dengue cases, NEA rainfall and temperature data, and URA planning-area boundaries.

The intended pipeline follows a Bronze-Silver-Gold architecture:

Ingestion — Retrieve live APIs and historical CSV/GeoJSON datasets through notebooks and scheduled jobs.
Bronze — Store raw source data in Delta tables.
Silver — Clean, standardise and spatially join weather stations and dengue activity to planning areas.
Gold — Generate modelling features for each planning area, including 7-, 14- and 28-day rainfall totals, rainy-day frequency, mean temperature, temperature anomalies and recent dengue activity.
Machine learning** — Train and evaluate models using time-based validation, with experiments tracked in MLflow.
Dashboard — Present predictions through a Databricks SQL / AI-BI interactive risk map.

The predictive model will estimate risk approximately 14 days ahead:

Risk_{t+14}=f(Dengue_t,\ Rainfall_{t-k:t},\ Temperature_{t-k:t},\ Location)

To make sure the model provides genuine value, its performance will be compared against a simple recent-history moving-average baseline instead of judging the model in isolation.