Silverwatch

Inspiration

Singapore officially became a super-aged society in 2026 (NPTD, 21.4% of citizens 65+), and social isolation among seniors is a known but under-measured risk. Through community programme work, we've visited multiple Active Ageing Centres and seen firsthand how small teams can be stretched thin serving many seniors, with newer centres often better-resourced than older ones. Outreach today is largely manual. We wanted to make prioritisation measurable.

What it does

SilverWatch ranks Singapore's towns by how many seniors living alone each Active Ageing Centre has to serve, today and projected to 2030. It surfaces a ranked outreach priority list for coordinators and volunteers, and simulates where new centres would relieve the most strain, using a greedy placement algorithm with a proven approximation guarantee.

How we built it

We pulled and cross-checked open datasets from data.gov.sg and SingStat: HDB elderly population by town, national living-alone rates by age group, MOH's Eldercare Services locations, and URA's planning area boundaries. We compared 5 forecasting methods head-to-head (naive, moving average, linear trend, Holt's exponential smoothing, and damped-trend Holt's) and backtested each one honestly before picking a winner.

Challenges we ran into

The dataset the challenge guide named for Active Ageing Centre locations turned out to be a national summary table, not a list of locations. We had to find the real one (MOH's Eldercare Services) ourselves.
We caught a real data bug: one town (Hougang) was split across two rows in the HDB dataset, which would have understated its senior population if not summed correctly.
We nearly shipped a "properly tuned" forecasting model that had actually overfit a tiny validation set, it performed worse than doing nothing. We caught it before it went on a slide.
With only town-level data available, we scoped our recommender to towns rather than exact sites, and were upfront about that limit rather than overselling precision we don't have.

Accomplishments that we're proud of

Every number in our submission is either independently verified or clearly labelled as an estimate, nothing is asserted without a source. We also grounded the problem in direct, firsthand experience with the centres themselves, not just secondhand data.

What we learned

With small datasets, simpler, well-validated models can beat more "sophisticated" ones, and honest reporting of a model's limits is more convincing to reviewers than overclaiming precision.

What's next for Silverwatch

With URA land-use and zoning data, we could move from recommending towns to pinpointing exact sites for new centres. We'd also like to add block-level, walking-radius coverage if finer-grained data becomes available.