Author: Nyashadzashe Nziramasanga
Nziramasanga, Nyashadzashe, 2021 Elderly falls prediction and prevention using machine learning, Flinders University, College of Science and Engineering
Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact copyright@flinders.edu.au with the details.
Advances in computational processing and the availability of large datasets in healthcare presents an opportunity to apply machine learning (ML) techniques to aid in predicting patterns in clinical datasets, assisting in diagnosing and treating patients. The thesis examines the application of three ML algorithms on aged care datasets to predict the severity of an elderly fall. The thesis also aims to investigate the appropriate means of monitoring, collecting data, and analysing the likelihood of elderly falls to reduce the costs related to elderly falls and the severity of falls in aged care facilities. Using falls incident reports and clinical information datasets sourced from AnglicareSA, a not-for-profit aged care and social services provider. Three ML algorithms were built for the research, namely Decision Tree Classifiers (DTC), Multi-Layer Perceptron (MLP) and Support Vector Machine (SVM). The models were compared against each other based on their accuracy, precision, recall and F-score to classify the severity of fall incidents and build a fall prediction model supporting clinical reasoning. About 2187 falls incidents and clinical data records remained after pre-processing. The mean age was 79, with most fall incidents were reported happening in bedrooms with minor outcomes for the severity of the fall. The accuracy for DTC, MLP and SVM was moderate at best recorded as 60%, 69% and 39%, respectively. The top five features that contributed significantly to predicting the severity of falls were incident location, age, number of incidents, facility, and respiratory rate. Even though the study explored the use of DTC, MLP and SVM algorithms to classify the severity of falls based on the recorded falls incidents and clinical health information, with reasonable prediction accuracy. However, future work is required to improve the accuracy of the ML models by using larger datasets of elderly falls and clinical datasets and applying wearable devices to help predict a fall.
Keywords: Decision Tree Classifier, Support Vector Machine, Multi-Layer Perceptron, Predictive modelling, Machine learning and Elderly falls
Subject: Computer Science thesis
Thesis type: Masters
Completed: 2021
School: College of Science and Engineering
Supervisor: Dr. Trent Lewis