We collaborate with the UCLA to develop a human behavior sensing and modeling platform for just-in-time anomalous human behavior interventions for depression, by means of smartphone and wearable sensor instrumentation and applied machine learning.
SENSATE: Entropy-AWARE Instrumentation for Just-In-Time Anomalous Human Behaviour Interventions
Notwithstanding the increased public awareness and the availability of effective pharmacological and cognitive behaviour therapies, Depression affecting many individuals goes undetected and untreated. The ability to detect the early onset of Depression in individuals will have significant impact on addressing depression. The current advancement in smartphone sensing, Experience Sampling Method (ESM) and human behaviour modelling using affective computing, multi-modal data fusion and machine learning algorithms has enabled the instrumentation smartphones to unobtrusively detect depression in individuals beyond laboratory confinements.
Given the complexity, heterogeneity of depressive behavior in individuals, and the scarcity of labelled objective depressive behavioral dataset, we borrow from methods in anomaly detection, entropy analysis and data quality metrics to enable just-in-time interventions for depressed or high-risk individuals .
Anomaly detection finds patterns in data that do not conform to expected behaviour. These nonconforming patterns are often referred to as anomalies, outliers, discordant observations, exceptions, aberrations, surprises, peculiarities, or contaminants. In contrast to standard classification tasks, i.e., traditional machine learning, anomaly detection can be applied on unlabelled data, taking only the internal structure of the dataset into account. The importance of anomaly detection is due to the fact that anomalies in data translate to significant, and often critical, actionable information in a wide variety of application domains, e.g., an anomalous MRI image may indicate the presence of malignant tumours.
On a longitudinal scale, entropy measures how order (i.e., sequence of events) deteriorates with the passage of time. We investigate the applicability of entropy analysis to observe how behaviour such as daily routines, activities, actions and social interactions deteriorate over time, a good indicator for depression.
WP1 : Instrumentation
Using AWARE, we developed Me, an android based application that unobtrusively collects smartphone sensor data that proxies objective bio markers proven in literature as promising in understanding depressive behavior. Me unobtrusively and passively collects contextual smartphone sensor data shown in the figure below, without interrupting or prompting the user. In addition to the sensor data, the Me also collects standard depression and personality test surveys for example Patient Health Questionnaire (PHQ-9), Beck Depression Inventory (BDI), and BIG-5 surveys. Me shows a notification to remind users when self reported survey are need due to be answered.
Me is privacy-aware and does not collect any personal information such as text, voice, telephone numbers, content of applications or web pages visited, except for meta data; for example, the time the screen was locked or unlocked, what application was launched at what a time, and when messages were received.
We conducted a one month long study with 4 participants ( 3 Females, 1 Male; 1 clinically diagnosed with Major Depressive Disorder.) . This study has provided us with valuable data to test the development of Anomaly detection , and entropy analysis methods for WP2. The study has also provided valuable insights for the improvment of the Me data sensing application
- Kennedy Opoku Asare, Aku Visuri, Denzil ST Ferreira. 2019. Towards Early Detection of Depression through Smartphone Sensing. In Proceedings of the 2019 ACM International Joint Conference and 2019 International Symposium on Pervasive and Ubiquitous Computing and Wearable Computers (UbiComp '19) . ACM. (To Appear)