Data visualisation and data mining technology for supporting care for older people
Nubia M. Gil, Nicolas A. Hine, John L. Arnott, Julienne Hanson, Richard G. Curry, Telmo Amaral, Dorota Osipovič · 2007 · Proceedings of the 9th International ACM SIGACCESS Conference on Computers and Accessibility (Assets '07) · doi:10.1145/1296843.1296868
Summary
This Assets '07 paper from a UK consortium (Dundee, UCL, Imperial College) reports on a telecare pilot study that instrumented the flats of older residents with a modest set of unobtrusive sensors — passive infrared motion detectors, pressure sensors, door contacts, electrical-appliance sensors — and used data warehouse, OLAP and data-mining techniques to look for changes in domestic 'busyness' that might signal changes in well-being. The authors position the work against a looming demographic and fiscal crisis: projected costs of Scottish residential and community care rising 85-155% between 2004 and 2019, with formal care workforce capacity unable to keep up. If relatively cheap ambient sensing can surface early warnings of decline, the argument goes, more people can be supported at home for longer and carers can be better targeted. The paper's methodological contribution is to deliberately avoid the more privacy-invasive approach of tracking specific Activities of Daily Living (bathing, dressing, feeding) via video or wearables, and instead use a coarser-grained measure — 'busyness' — that simply counts sensor firings per room per time zone and looks at their trends. Data from one flat over nine months is analysed as a case study. The resident had non-insulin-dependent diabetes and underwent two medication changes during the study (weeks 9 and 16), giving a natural experiment to check whether the busyness signal tracks clinically meaningful events. OLAP cubes are used to visualise sensor activity at different granularities (day, week, time-of-day zones like 'lunch' or 'late evening'); Weka-based C4.5 decision trees are used to extract rules characterising normal patterns, against which deviations could later be flagged.
Key findings
Busyness trends tracked the medication changes: increased evening activity around the living room lamp (the resident did crosswords and Sudoku there), a moderate decrease in bed-sensor firings after week 15, and fluctuations in hall and bathroom activity that clustered around the medication-change weeks. Trend lines reached correlation factors (R²) around 0.11-0.25, which the authors treat as signal worth escalating to human carers rather than as evidence for automated alerts. The decision-tree rule-mining produced a set of seven interpretable rules classifying time-of-day from room-level sensor counts (for example: 'living room <=0 AND bed 1-8 -> late evening') with 86.8% training accuracy and 72.6% test accuracy. The authors are candid that older residents are *not* always as regular as the activity-monitoring literature often assumes; the most stable patterns were around sleep, meals, and scheduled telephone calls, and the researchers argue future work should relax hard rule thresholds in favour of fuzzy limits. They explicitly frame the system as supporting a 'dialogue of care' between resident and carer rather than replacing clinical judgement, and treat privacy as a first-order design constraint — busyness preserves privacy better than ADL-level monitoring because it does not attempt to identify specific activities.
Relevance
Seventeen years on, the privacy-respecting ambient-sensing agenda outlined here is still live — in commercial form via devices like smart plugs, motion-based fall detection, and 'aging in place' platforms, and in research via passive wellbeing monitoring for people with dementia, depression, or chronic conditions. The paper is a useful historical reference for anyone working on home-based care technology or remote monitoring for disabled users, precisely because it frames the ethical trade-offs explicitly: the authors chose coarse busyness counts over ADL-level inference because the latter trades privacy for specificity, and that debate has only intensified with more capable AI. The observation that 'older people's lives are not very regular' is a useful corrective to the many systems that assume detectable deviations imply decline. Limitations are real — a single-flat case study, modest R² values, and no evaluation of whether the surfaced trends actually improved carer decision-making or resident outcomes — but the paper's framing of technology as supporting *dialogue* rather than automating judgement is worth preserving as a design principle.
Tags: aging · older adults · independent living · telecare · ambient sensing · data mining · data visualisation · activities of daily living · lifestyle modelling · privacy · home-based care · smart home