Inspiration

We were inspired by a simple question, what if social workers could know which communities need help the most before those needs become a crisis? Currently, outreach by social-care organisations might be manual and uneven. We just wanted to explore how publicly available data could be brought together to help social workers and volunteer coordinators identify communities that may need greater attention and prioritise limited resources.

What it does

Our project is a data-driven decision-support concept designed to identify communities that may be at higher risk of social isolation. It combines demographic, living-arrangement and community-support data to generate an area-level Community Isolation Risk Score. Instead of only showing statistics, we aim to answer three practical questions where are the communities that need the most attention? why are they at higher risk?what type of outreach or community support should be prioritised? The intended user experience follows: Identify → Explain → Prioritise → Act. The user can explore a map of Singapore, select a planning area, view the risk level and its contributing factors, and receive recommended outreach priorities. The current concept operates at the community/planning-area level, rather than attempting to identify individual seniors or real-time emergencies.

How we built it

Our planned pipeline follows a simple data-to-decision workflow: Open data → Ingestion → Transformation → Risk scoring → Dashboard. We plan to combine datasets including: HDB Elderly & Future-Elderly Population, Resident Population 65+ by Living Arrangements, Senior Activity Centres & Active Ageing Centres and Households Assisted Through ComCare Schemes. The data will be ingested into Delta Lake, cleaned and joined into analysis-ready tables, and used to construct an area-level risk score based on indicators such as elderly population density, living-alone rates and community-support coverage. We plan to use MLflow to track model experiments and Databricks AI/BI to present the results through an interactive dashboard. Unity Catalog can be used as a governance layer to document datasets and lineage. This follows the recommended Databricks workflow of ingestion, data engineering, machine learning, analytics/BI and governance.

Challenges we ran into

One of our biggest challenges was making sure our idea matched what the available data could actually support. Our initial vision involved identifying seniors who might need immediate help, such as someone living alone who has fallen and cannot seek assistance. However, the available datasets are primarily aggregated at the planning-area, town or community level. They cannot directly tell us that a specific individual has fallen or currently needs food or medical assistance. We therefore refined our scope to focus on community-level vulnerability and outreach prioritisation. Rather than claiming to predict individual emergencies, SilverLinings identifies areas where social-care resources may be most needed and explains the factors contributing to that risk. Another challenge was designing a risk score that is useful without making it appear more certain than the underlying data allows. We want the output to support human decision-making, not replace the judgement of social workers.

Accomplishments that we're proud of

We are proud that we turned several fragmented public datasets into a single, actionable concept. Instead of creating another dashboard that simply displays elderly population statistics, we designed SilverLinings around the decision a community worker actually needs to make: where should we act first? Our concept also aligns closely with the intended A1 challenge: a geospatial view of planning areas, a composite risk score, prioritisation of high-risk areas and a dashboard designed for social workers or volunteer coordinators. Most importantly, we learned to distinguish between what our data can support today and what could become possible with additional authorised data in the future.

What we learned

We learned that building a useful data product is not just about finding datasets and training a model. The most important part is connecting the data to a real decision. We also learned the importance of understanding the granularity and limitations of data before designing a solution. A model can only make claims that its underlying data supports. Through this challenge, we are also learning how to connect the different stages of a data project — from ingestion and transformation, to modelling, experiment tracking, visualisation and governance — into one reproducible Databricks workflow.

What's next for DAISI challenge (SilverLinings)

Our next step is to turn the concept into a working MVP.

We plan to:

Ingest and clean the selected datasets. Build the area-level Community Isolation Risk Score. Validate the score against simple baseline approaches. Create an interactive map showing risk levels across planning areas. Add explanations showing the main factors contributing to each area's risk. Develop actionable outreach recommendations. Track experiments using MLflow and incorporate appropriate governance through Unity Catalog.

Our longer-term vision is to move beyond identifying where communities need support towards understanding how resources can be allocated most effectively. With additional authorised and appropriately governed data, SilverLinings could eventually support more granular case prioritisation and resource allocation.