Senior Data Science: Safe Aging with SPHERE

Contribute to open, cutting-edge research on the use of wearables in promoting health and independence for seniors. #health

$10,000 in prizes
jul 2016
577 joined

One of the biggest impediments to independent aging-in-place is the ability to respond when something goes wrong. By combining passive sensors like wearables with machine learning algorithms, modern technology can offer unobtrusive, private ways to monitor activities and learn when response is needed.

Why

Just in the US, more than 10 million adults are 65+ years of age and live alone. One of the biggest impediments to independent aging-in-place is the ability to detect and respond when something goes wrong. We now have the ability to use passive sensors (like motion detectors and accelerometers) to monitor some aspects of an older person’s activity privately and unobtrusively. But we don’t yet have reliable ways to turn this information into an actionable view of what is happening to them in real life.

The Solution

The SPHERE Inter-disciplinary Research Collaboration assembled a unique dataset in their research home in Bristol, UK. In addition to recording the outputs from smart technologies like wearables and environmental sensors, the center also has human-labeled descriptions of what seniors were doing—their activities, like meal preparation or watching TV; and their positions, like walking or sitting—while those technologies were in use. Competitors were charged with building computer algorithms that could translate the easierto- collect sensor data into harder-to-collect information about real activities.

The Results

Of the nearly 1000 submission generated during the competition, the winning models were able to make that translation correctly (predicting what seniors were doing as described by human researchers) with ~90% accuracy, a hugely impressive result. This performance improved ~35% over the previous benchmark and surpassed even internal state-of-the-art models that the researchers had developed.


RESULTS ANNOUNCEMENT + MEET THE WINNERS

WINNING MODELS ON GITHUB