Künstliche Intelligenz für die Optimierung von stadienspezifischen Trainingsplänen für Demenzpatient*innen

Abstract

Dementia, particularly Alzheimer’s disease, affects millions of people worldwide. Currently, no cure for Alzheimer’s is known, but the progression of the disease can be counteracted. To this end, there are a number of medicinal and non-medicinal approaches
that have already achieved initial successes. MAS Alzheimer’s Aid is a non-profit organization specializing, among other things, in conducting Alzheimer’s training. Trainers
create training plans and conduct sessions where individuals with Alzheimer’s disease
can exercise their physical and mental abilities. When planning these time-consuming
training sessions, trainers must ensure that the exercises planned will spark high interest
among the participants.
As part of this master’s thesis, a system was developed that uses machine learning
to automate the creation of optimized training plans. During the creation of these plans,
it is essential to determine whether potential exercises will generate high interest among
participants or not.
For this purpose, a comprehensive analysis of the database was initially conducted
using statistical methods, providing detailed insights into the structure and success of
previously conducted training sessions. Building on this, machine learning models were
developed to evaluate the suitability of an exercise for a planned training session. The
focus was on identifying potentially unsuitable exercises and excluding them from future
sessions. To identify the best model, a series of unsupervised and supervised classification
algorithms were applied. No clear separation between good and bad exercises could be
identified using unsupervised learning methods. In contrast, through the application
of various resampling strategies, testing different feature combinations, and extensive
hyperparameter optimization for supervised learning algorithms, a suitable model was
developed. This model can accurately predict whether an exercise is well-suited for a
planned training session or not.
The results achieved mark an important step towards the forthcoming integration of machine learning models into the TrainYourBrain Optimizer, which will enable
Alzheimer’s trainers to digitally and automatically create optimized training plans.

Date of Award	2024
Original language	German (Austria)
Supervisor	Stephan Winkler (Supervisor)