Use of historical data to improve sensor accuracy and prioritize signals
The 900 clients, of whom multiple disabled patients, of a care institution in the Netherlands are monitored intensively. During the night 5 watchmen receive some 5000 notifications on average per night. Notifications come from different sensor: movement, sound, doors. Key question was to get more insight into the different types of notifications in order to prioritize these.
We collected the historical notifications caused by the various sensors. We first analysed which sensors are malfunctioning as well as which notifications are false positives.
Finally we identified reasons underlying each notification: noise (e.g. a train passing by, lightning), sensor sensitivity, etc. and we identified the true positives
We developed a traffic light approach to prioritize the notifications as some notifications require particular action by the watchmen. In addition we adjusted sensor information to reduce noise (false positives)