Data synthesis and classification through machine learning for detecting mild cognitive impairment
| dc.contributor.author | Aljumaili, Mahmood | |
| dc.contributor.examiningcommittee | Jeffrey, Ian (Electrical and Computer Engineering) | en_US |
| dc.contributor.examiningcommittee | Eskicioglu, Rasit (Computer Science) | en_US |
| dc.contributor.supervisor | Friesen, Marcia (Faculty of Engineering) | |
| dc.contributor.supervisor | McLeod, Robert | |
| dc.date.accessioned | 2022-03-31T15:07:33Z | |
| dc.date.available | 2022-03-31T15:07:33Z | |
| dc.date.copyright | 2022-03-30 | |
| dc.date.issued | 2022-03-30 | |
| dc.date.submitted | 2022-03-30T19:34:02Z | en_US |
| dc.degree.discipline | Electrical and Computer Engineering | en_US |
| dc.degree.level | Master of Science (M.Sc.) | en_US |
| dc.description.abstract | This thesis examines Machine Learning (ML) applicable to detecting Mild Cognitive Impairment (MCI) from gameplay data of a Serious Game (SG). The foundation of the work is a newly developed SG called WAR Cognitive Assessment Tool (WarCAT). WarCAT is based on the familiar card game WAR and is played on mobile devices (phones, tablets). WarCAT captures players’ moves during the game to allow inference of cognitive processes of strategy recognition, learning, and memory. This thesis focuses on the potential of detecting MCI by developing a data synthesis model and a classifier model to be applied to WarCAT gameplay. The data synthesis model generates WarCAT gameplay data similar to that of a human, so that it can be used to train and validate the classifier model. The classifier model has the potential to detect MCI from a player’s gameplay data. Machine Learning (ML) methods are used by both models. The classifier uses a deep Artificial Neural Network (ANN) that is first trained on a large labelled dataset to help detect MCI. The data synthesis model generates synthetic data to plausibly emulate a large population of players. The sub-paradigm of ML, namely Reinforcement Learning (RL) is used to generate synthetic data, as it most closely emulates the way humans learn. An RL bot undergoes millions of sessions of trial and error (training), processing millions of gameplay training patterns and achieves results comparable or better than the best human performance. A conjecture is that an RL bot with fewer training opportunities emulates a person with less experience or potentially declining cognitive function. This baseline allows for the creation of bots with the goal to emulate individuals at various stages of learning, or conversely, various levels of cognitive decline. The thesis demonstrates the ML framework necessary to generate and classify different levels of play, which correspond to different levels of players’ cognitive states or abilities. | en_US |
| dc.description.note | May 2022 | en_US |
| dc.identifier.uri | http://hdl.handle.net/1993/36375 | |
| dc.language.iso | eng | en_US |
| dc.rights | open access | en_US |
| dc.subject | Mild Cognitive Impairment | en_US |
| dc.subject | Serious Games | en_US |
| dc.subject | Machine Learning | en_US |
| dc.subject | Reinforcement Learning | en_US |
| dc.title | Data synthesis and classification through machine learning for detecting mild cognitive impairment | en_US |
| dc.type | master thesis | en_US |