SilverWatch

SilverWatch: Proactive eldercare through block-level intelligence Inspiration Singapore becomes super-aged in 2026, when 1 in 5 residents will be 65 or older. That demographic shift includes 88,400 residents aged 65 and above living alone, which is 11.5% of all seniors and more than double the number from a decade ago. Social isolation predicts early mortality more strongly than obesity or smoking, but isolated seniors remain hard to find.

Current outreach relies on manual referrals and whoever walks into an Active Ageing Centre (AAC). Active seniors stay connected, while isolated seniors in rental blocks often go unnoticed until a crisis. Town-level averages hide these local blindspots, as a cluster of rental blocks can sit inside an otherwise well-served planning area without standing out. Volunteer hours spent visiting well-served blocks mean fewer visits for seniors who have no support at all.

We designed SilverWatch to replace local estimates with block-level data clustering, replacing reactive outreach with targeted planning.

What it does SilverWatch is a spatial analytics framework designed for two groups:

HDB and MSF city planners: Evaluates and simulates proposed AAC sites on an interactive map. Planners can place a proposed site to measure walking coverage within 400 meters, tracking changes in AAC capacity per 1,000 seniors and the immediate drop in area-wide isolation risk.

Social workers and volunteer coordinators: Replaces guesswork with a prioritized block-level outreach list. It pairs walking routes with dominant home language and dialect profiles to send appropriate volunteers, while identifying nearby Community Clubs or AACs for follow-up care.

Instead of working as a static dashboard, SilverWatch updates dynamically. Risk scores recalculate when source datasets refresh, volunteer visit logs feed back to show where coverage lapses, and proposed AAC locations update community impact immediately.

How we built the concept [SingStat & HDB Data] -> [Feature Normalization] -> [Hierarchical Clustering] -> [Risk Score S(b)] | v [Actionable Routes & AAC Sites] <- [400m Walking Radius] <- [Spatial Mapping & AAC Simulation] Our pipeline uses open data from SingStat (Key Indicators on the Elderly), HDB (Elderly & Future-Elderly Population), and MSF (Senior Activity Centres & AAC Locations).

Data aggregation and feature engineering For each residential block b, we model three normalized features:

P(b): Density of elderly (65+) and future-elderly (55–64) residents.

L(b): Proportion of living-alone proxy indicators (rental units and single-member households).

D(b): Pedestrian distance to the nearest active AAC.

Hierarchical clustering model To group vulnerable communities without imposing arbitrary boundaries, we apply Agglomerative Hierarchical Clustering using Ward’s minimum variance method. Ward's criterion minimizes total within-cluster variance when evaluating merges:

ΔESS(A,B)= ∣A∣+∣B∣ ∣A∣⋅∣B∣ ∥μ A −μ B ∥ 2

Where μ A and μ B represent feature centroid vectors for clusters A and B.

Dynamic isolation risk index We calculate a composite Isolation Risk Score S(b) for each block b:

S(b)=w 1 ⋅P(b)+w 2 ⋅L(b)+w 3 ⋅D(b) Subject to ∑w i =1. When planners place a proposed AAC site k at coordinates (x k ,y k ), distance D(b) updates:

D new (b)=min(D old (b),dist(b,k)) This recalculates S(b) immediately, giving planners a clear measure of risk reduction (ΔS community ).

Challenges in the conceptual phase Town-level statistics vs. block-level granularity: Open data often aggregates statistics at planning area levels. Reconciling macro demographic indicators down to micro HDB block estimates without introducing bias required careful spatial normalization.

Pedestrian distance metrics: Straight-line radius metrics do not reflect mobility for seniors. Using a 400-meter walking radius ensures site simulations reflect actual physical accessibility instead of Euclidean distance.

Usable output design: Turning cluster outputs into clear decisions for field teams required structuring technical outputs into simple formats, such as language-matched volunteer routes.

What we learned Proximity does not equal connection: High-risk senior clusters can exist right next to major transport nodes or commercial hubs, because macro-level data obscures localized isolation.

Closed-loop design matters: Static dashboards quickly go out of date. Designing the system around continuous feedback (such as volunteer logs updating risk scores) keeps the data relevant over time.

Next steps for SilverWatch Prototype implementation: Build the core spatial engine and interactive map interface.

Volunteer feedback integration: Create schemas for logging ground visits to refine real-time risk re-scoring.

Dialect routing optimization: Implement automated volunteer-to-block matching based on localized language profiles, including Hokkien, Teochew, and Cantonese.