Inspiration
Where should we act first?
Singapore already publishes good data on where seniors live and how neighbourhoods are ageing. But outreach to isolated seniors still usually starts from one flag: does the senior live alone?
That flag misses most of the people it is meant to find. In a national survey of Singaporeans aged 60+ (THE SIGNS-I, 2016–17), 33.9% felt lonely at least sometimes, but only 7.3% lived alone. Even if every senior who lived alone were lonely, nearly 4 in 5 lonely seniors would still live with others.
It matters for health too. People with strong social ties have about 50% higher odds of survival, an effect comparable to quitting smoking (Holt-Lunstad et al., 2010).
SilverWatch turns scattered ageing and social-care data into one clear view of where loneliness risk is concentrated, why, and what help fits.
See the risk. Understand the driver. Take the right action.
What it does
SilverWatch is a decision-support tool for social workers, volunteer coordinators and planners, built around an interactive map of Singapore's planning areas.
Each of the 30 planning areas reported separately in the Census gets a LoneliScore. It is the predicted relative risk that residents aged 65+ feel lonely, against the Singapore average:
- 100 = Singapore average
- 110 = 10% above average
It is a prediction from open data, not a measurement. SilverWatch scores areas, never people.
- See the risk. A heat map shades each area by LoneliScore. A top-10 leaderboard lists the areas to reach first: Red, then Amber, each ordered by estimated lonely seniors. Today's top 10 hold about 4 in 10 of Singapore's seniors.
- Understand the driver. Hover over an area to see its score, likely range and what pushes it up or down, in points: living alone, widowhood, sight or hearing, walking or self-care, and memory or communication.
- Take the right action. Each area gets an action tier:
| Tier | When | What happens |
|---|---|---|
| 🔴 Red: reach out now | Consistently in the top fifth of areas and 10%+ above average | Silver Generation Office home visits (rental and 1–2-room blocks first), extra befriending |
| 🟠 Amber: screen and strengthen | Consistently above average | Proposed 3-question loneliness check at centres and home visits, weekly group activities |
| 🟢 Green: keep universal | All other areas | Routine Active Ageing Centre (AAC) programmes |
The main driver tailors the action:
- Memory or communication points to dementia screening. It is the main driver in 5 of the 8 Amber areas.
- Widowhood points to peer and bereavement groups.
- Sight points to Project Silver Screen.
What the prototype already shows:
- Head count is not risk. Bedok had the most seniors in Census 2020 (53,370) but scores an average 99. Outram had only 4,680 but scores the highest, 119.
- Living alone is not the whole story. Woodlands has fewer seniors living alone than average, yet scores 109, driven by difficulty remembering, communicating and walking. A living-alone map would rank it low.
👉 Try the live prototype: LoneliScore Explorer
How it meets the A1 brief
| The brief asks for | SilverWatch |
|---|---|
| Heat map of planning areas | LoneliScore by planning area |
| Risk index | Living alone, widowhood, age and six kinds of difficulty, each weighted by how strongly it goes with loneliness |
| Elderly density | Number of seniors turns risk into workload (estimated lonely seniors), which orders the top 10 |
| AAC coverage gaps | No open, current list of AAC locations exists. The only open map is a 2016 snapshot with no AACs, against 230+ centres today. AAC reach is built as a separate layer, ready when a current list arrives |
| Top 10 and a usable dashboard | Both in the working prototype |
How we built it
- Learn. No open Singapore survey records loneliness person by person, so we learned the weights from the open US National Health Interview Survey 2024: 10,123 adults aged 65+, of whom 475 feel lonely "usually or always". The model uses only traits Singapore publishes, or can estimate, by planning area:
- age and sex;
- widowhood and living alone;
- six kinds of difficulty: seeing, hearing, walking, remembering, self-care and communicating.
- Check. We tested on people the model had not seen, using 5-fold cross-validation in the US survey, and picked as many seniors as live alone (28%). The score's picks included 68% of those who felt lonely, against 57% for the living-alone flag.
- Apply. Each area's Census 2020 profile of residents aged 65+ goes through the same weights:
$$\text{LoneliScore} = 100 \times \frac{\text{predicted share lonely in the area}}{\text{predicted share lonely in Singapore}}$$
Estimated lonely seniors use Singapore's own measured level: 12.1% "mostly lonely" among those aged 60+ (THE SIGNS-I).
- Rank with honesty. We reran the model 1,000 times on resampled survey data to give every area a likely range. "Consistently" means in at least 80% of reruns. Areas with fewer than 5,000 seniors, many young single residents or a wide range get a "check locally" badge.
- Show it. Today the pipeline runs in Python and feeds the interactive explorer. In the two-week build it moves onto Databricks:
- Bronze → Silver → Gold Delta tables;
- the model tracked in MLflow;
- an AI/BI map dashboard;
- a Genie Agent for plain-English questions;
- area briefs written with Foundation Model APIs;
- access governed in Unity Catalog.
Challenges we ran into
- Data at different levels. Some tables are national, some are by planning area. Living alone and widowhood are not published by area for seniors. We estimate them from national rates by age, adjusted for each area's age mix and its share of one-person households or widowed residents, and we label them as estimates.
- No current open list of Active Ageing Centres. We built AAC reach as a separate layer that decides how help is delivered (walk-in or outreach), and we are asking the organisers for a current list.
- Defining loneliness responsibly. Living alone is not the same as being lonely, and public data cannot see feelings. LoneliScore is a predicted relative risk for an area, never a label on a person.
- Being honest about certainty. Some tiers depend on how living alone is estimated. We scored every area under three estimates. The tool shows which tiers hold under all three (17 of 30) and which to confirm locally first.
Accomplishments that we're proud of
- A working prototype on real open data, not mock numbers. All 30 planning areas are scored: 4 Red (Outram, Geylang, Toa Payoh, Bukit Merah), 8 Amber and 18 Green. The Red tiers depend on how living alone is estimated, so they are marked for local confirmation.
- A score that beats the simple rule. In US test data it found more lonely seniors than the living-alone flag at the same list size (68% vs 57%).
- More than a heat map. Every area shows what drives its score and a suggested action tied to real Singapore programmes.
- Risk kept separate from workload. The leaderboard puts risk first (tier), then workload (estimated lonely seniors).
- Designed for non-technical users. Open the map, see the priority areas, read a plain recommendation. No statistics needed.
What we learned
- Open data can rank areas but cannot see feelings. A short, consented loneliness check at centres is what will turn a US-trained score into a Singapore one.
- Showing uncertainty builds trust. A likely range and a "check locally" badge help a planner more than a single number that looks precise.
- Living alone is a weak guide on its own. The areas that most need attention are not always the ones with the most seniors, or the most seniors living alone.
What's next for SilverWatch
- Make it Singaporean. Add consented 3-question loneliness checks at AAC and Silver Generation Office visits to recalibrate the model. Then replace the US weights with representative Singapore survey data (for example THE SIGNS or WiSE).
- Add AAC reach. With a current centre list, fill the coverage-gap layer and add an AAC siting tool.
- Resource allocation and what-ifs. Enter volunteers, outreach hours or mobile teams and see where they do the most good. Test what happens if a new AAC opens.
- Look ahead. Project how risk shifts to 2030 as the population ages.
SilverWatch helps Singapore move from seeing loneliness risk to acting on it: earlier, smarter, and where it matters most.
Built With
- claude
- codex
- python
