There are a sequence of events that will complete the whole system analytics.
Here is a visual representation of the whole system.
Here's a run through of the main system elements:
|Data stream||A data stream is an organisational unit for data about a single person that comes from a single source. For example, the data reported by a PIR sensor in the kitchen for Entity A would be one data stream and the data reported by a PIR sensor for the same person from the sitting room would be a different data stream.|
|Raw data||Raw data is data uploaded by the client.|
|Processed Data||Processed Data are from the raw data after processing. They are ready to be analysed and are typically well structured with a pre-set sampling rate dependent on the data type. For example, PIR data might be processed down to show number of firings every 5 minutes.|
|Behaviour pattern||A pattern is calculated from several days of processed data and shows what is ‘normal’ in terms of the data reported by a data stream.|
|Indicators||Indicators are calculated daily from Processed Data per Data Stream per entity ID. An indicator shows how normal a given day’s worth of data is compared to the pattern for that data stream.|
|alerts||Alerts are calculated for a data stream and show when data for that data stream are outside the normal parameters according to the pattern.|
|Statistics sets||A statistics set is a set of secondary statistics calculated for indicators that help in the understanding of the indicator data.|
The data fusion step in the CareTec.AI data processing model incorporates the automatic weighting of the indicators relating to a particular entity. And the fusion of all indicators to create an overall indicator, as well as a statistic set and alerts for the overall indicator.
The diagram below shows the processing of data from a PIR motion sensor from raw data through to an identified pattern and into an individual indicator. The data shown is the activity value recorded from the sensor over an entire day over a 70 day period. The data is very noisy and patterns in the data are difficult to identify by eye.
The next images shows the behavioural pattern identified from the PIR motion data. The pattern detection is based on a probabilistic machine learning model and the graphic shows the identified pattern in a probabilistic way.
For each time point in the pattern, there is a probabilistic interpretation of how likely a given motion count is, where a given motion count is very likely the patterns shows green, where it is highly unlikely the pattern shows red.
The pattern shown below shows an increase in probability of higher motion counts around 08:00, a dip around midday and a further increase in the evening.
It should be noted that the graphic displayed here is only one way of interpreting data in the model that we use and other representations of a normal pattern can be developed to suit the needs of a client.
The following diagram shows the individual indicator for the PIR data, which shows how normal a given days data is based on the probabilistic pattern that has been identified.
The higher the score, the less likely (and therefore the less normal) the data for the given day. Green-Amber-Red thresholds are calculated based on the data and allow for easy traffic-lighting of the data.
The next diagram shows the multi-sensor data analysis and fusion approach that CareTec.AI uses across a sample of sensor data.
The next image shows the Primary or Overall Indicator for the sample data. The interpretation of this graph is the same as that of an individual indicator, but applies to the entirety of the data-set.