Detecting and tracking behavioural patterns through Personalized Ambient Monitoring
J.D. Amor and C.J. James
Bipolar disorder (BD) is a serious mental disorder that affects between 0.4 and 1.6% of the population  and costs the UK economy £2 Billion annually. BD is characterized by recurring episodes of mania and depression. Mania typically presents with an elevated mood, decreased need of sleep and pressure of thoughts and speech. Conversely, depression typically presents with depressed mood, decreased energy and increased need of sleep. BD can be very disruptive to a person’s daily life, but can be controlled through pharmaceutical and therapeutic methods.
Many of the therapeutic treatments for BD focus on maintaining a stable daily rhythm and minimizing the stressful experiences that can trigger an affective episode. In addition to this many people with BD self-monitor their condition, keeping records of their activity and mood, in order to identify the onset of an affective episode so that steps may be taken to lessen the severity of the episode .
The Personalized Ambient Monitoring (PAM) project  is aiming to develop a self-help tool for monitoring BD that can track a person’s behavioural patterns and from changes in those patterns, predict the person’s mood state. To date, we have developed a system that monitors behavior in an ambient and unobtrusive manner and which can detect a person’s normal behavioral patterns and detect when their behavior deviates from this norm.
We use a number of discrete sensors to provide monitoring in the home environment and also in an ambulatory setting. In the home environment we have a bespoke environmental node  and additional sensors such as PIRs, light sensors and a camera . For ambulatory monitoring we use a custom built wearable node coupled to a mobile phone. The data gathered from these sensors are processed in a training period, using Continuous Profile Modeling (CPM) , to extract the underlying behavioral patterns. Once the behavioral patterns have been established, further data can be tested against these to detect changes.
In this work we give an overview of the PAM system, describe the sensors used and explain the data processing methodologies. Some results from a technical trial of the system, including a wearable component, are presented.