Inspiration

Singapore is ageing rapidly, but social isolation is difficult to see. Existing outreach can tell us who has already been reached, but it is harder to identify communities where seniors may be more vulnerable and where limited outreach resources should go next. That inspired Silver Lantern: not a system that claims to predict whether an individual senior is lonely, but a decision-support tool that uses observable community-level risk factors to help organisations find potential blind spots and prioritise outreach.

What it does

Silver Lantern combines demographic, household, socioeconomic and eldercare-access data to build an explainable vulnerability score across Singapore. It helps users answer three questions:

  1. Where? Which communities show greater structural vulnerability to social isolation?
  2. Why? What factors are driving each area's score?
  3. What next? Given limited outreach capacity, where should additional resources be prioritised?

Rather than stopping at a heatmap, Silver Lantern is designed to translate data into an actionable outreach allocation.

How we built it

We designed the project around Singapore's open government data ecosystem, using datasets from data.gov.sg, SingStat, HDB, MSF and MOH. Our proposed data model combines factors such as:

  • elderly and advanced-age population;
  • household structure and living arrangements;
  • difficulty performing basic activities;
  • household income and employment;
  • locations of eldercare services.

These inputs are transformed into standardised indicators and combined into an explainable composite vulnerability score. We plan to test alternative weighting assumptions to ensure that priority areas are not simply artifacts of arbitrary weights. The second layer is an outreach optimisation model. Given a fixed capacity—for example, a limited number of volunteer-hours—it recommends how those resources could be allocated across higher-priority communities. On Databricks, the planned pipeline is: Open Data → Lakeflow → Delta Lake / Unity Catalog → Analytics & Optimisation → SQL / AI-BI → Databricks App MLflow will support model/version tracking and sensitivity testing.

Challenges we ran into

Our biggest challenge was data granularity. Some datasets initially appeared suitable for geographic modelling but, upon inspecting the actual data, contained only national-level time series. We therefore avoided artificially turning national statistics into neighbourhood-level estimates. We identified additional official datasets—including Census 2020 population, household structure, basic-activity difficulty and household-income data by planning area, together with MOH's geospatial Eldercare Services dataset—to build a more defensible geographic model. A second challenge is conceptual: living alone is not the same as being socially isolated. Silver Lantern therefore measures structural vulnerability to isolation, rather than claiming to diagnose loneliness or predict an individual's social condition.

Accomplishments that we're proud of

We are particularly proud that Silver Lantern does not use AI or machine learning simply for the label. Where the data supports scoring, we use transparent scoring. Where the problem is resource allocation, we use optimisation. Where uncertainty exists, we make it visible. This makes Silver Lantern's recommendations easier to explain, challenge and improve—important when the output could influence where real community resources are deployed.

What we learned

The project reinforced a simple lesson: a sophisticated model cannot compensate for data that does not support the claim being made. Inspecting the underlying datasets changed our original approach. Instead of trying to “predict social isolation,” we reframed the problem around something open data can reasonably support: identifying community-level vulnerability and service-access gaps. We also learned that the most useful analytics product is not necessarily the one that makes the most predictions. Often, the more valuable question is: given what we know and what we have, what should we do next?

What's next for Silver Lantern

Our two-week MVP will focus on a Singapore-wide vulnerability model, interactive geographic view, explainable risk drivers and outreach allocation simulator. Beyond the MVP, we want to incorporate updated demographic data, improve service-access modelling using travel distance rather than simple proximity, validate the vulnerability weights with domain experts, and explore future ageing scenarios to identify where eldercare and outreach capacity may be needed most. Ultimately, Silver Lantern could evolve from identifying today's outreach blind spots into helping Singapore plan ahead for tomorrow's ageing communities.

Built With

Share this project:

Updates

Submission history