Anomaly detection with hierarchical temporal memory: a system for detecting potential distress in the elderly and those with dementia
Date
2020
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
As we age, it becomes more difficult to maintain our independence. This can
be due to the prevalence of dementia among the elderly, or the high probability of
developing disabilities as we age [1][2]. The loss of independence has been linked to
depression, suicide, somatic symptoms, and cognitive changes [3]. ☐ The goal of this research was to design a system that leverages machine
learning and wearable Internet of Things (IoT) technology to extend the independence
of older individuals and those with early stage dementia. Using these technologies, it
is possible to learn a user’s routines and detect anomalous activity that could indicate
distress, and, if necessary, notify a caretaker, guardian, or loved one. Using
Hierarchical Temporal Memory (HTM), I built a dual continuous, online,
unsupervised model for detecting anomalous events in streaming heart rate and
geospatial data. Using synthetic heart rate and temporospatial data, I was able to
achieve an area under ROC curve of 0.947 for the heart rate model, and 0.999 for the
geospatial model. Accuracy of 0.99 was achieved for both models with precisions of
1.0 and 0.5, for the heart rate and geospatial models respectively
Description
Keywords
Anomaly Detection, Dementia, Machine Learning, Depression, Cognitive issues, Wearable technology, Internet of Things