LE3 .A278 2021

2021

Silver, Danny

Acadia University

Master of Science

Masters

Computer Science

Computer Science

Patients with dementia will eventually experience a significant loss of cognitive function. As a result, they will have difficulty solving and communicating their physical problems and emotional pain appropriately. This leads to frustration that can manifest itself as lower-risk behaviours such as restlessness and repetitiveness to higher risk behaviours such as agitation and verbal or physical aggression. Therefore, monitoring the risk of a resident harming themselves or others is a priority within a long-term care facility where dementia is present. Caregivers are trained to recognize signs of degrading states of dementia and common sources of stress for people with dementia but predicting when the aggressive behaviour is going to happen needs more investigation. We investigate the use of structured data and unstructured text data from Electronic Health Records (EHR) to predict a residents future level of aggression by using natural language processing and machine learning techniques. The project involved significant work to anonymize patients records to meet Research Ethics Board (REB) requirements. A level of aggression (LoA) metric is created to provide a barometer of a resident’s aggressive behaviour and to overcome a spars collection of aggressive acts, even for the most aggressive resident. The project focuses on inductive decision trees so as to provide human readable models for analysis of important factors and their relation to level of aggression

The author retains copyright in this thesis. Any substantial copying or any other actions that exceed fair dealing or other exceptions in the Copyright Act require the permission of the author.

https://scholar.acadiau.ca/islandora/object/theses:3694