Determining the Severity of Dementia Using Ensemble Learning

• Published in: Big Data Analytics Vol. 13773; pp. 117 - 135
• Main Authors: Srivatsan, Shruti, Bamrah, Sumneet Kaur, Gayathri, K. S.
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer 2023; Springer Nature Switzerland
• Series: Lecture Notes in Computer Science
• Subjects: Activities of daily living; Dementia; Ensemble learning; Machine learning; Magnetic resonance imaging; Smart healthcare
• ISBN: 3031240936; 9783031240935
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-031-24094-2_8

According to WHO, 10 million individuals are subject to the risk of dementia annually. Geriatric care is expensive, restricted, and inaccessible for dementia care. The condition has no cure and medication is provided to inhibit the progress. Individuals exhibit a decline in cognition at severe stages impacting behaviors, hence detection of the early onset of dementia is critical. As an individual ages, the symptoms of dementia become harder to distinguish from symptoms of normal ageing. Behavioral patterns are identified using Activities of Daily Living (ADL) data. Traditional forms of dementia prediction widely use Magnetic Resonance Imaging (MRI). The detection of dementia and the severity utilizes both forms of data indvidually largely. A novel approach is proposed to combine motion sensor and brain scan data alternatively to detect the presence and eventually severity of dementia. Patterns of behavior are recorded, the interactions between the occupant and sensors placed in a smart environment. Daily activity recordings are used to detect the presence or absence of dementia in Phase 1. Phase 2 focuses on the utility of MRI data in longitudinal and cross-sectional form to further assess the severity. Distinct features are extracted using feature selection method. The hyperparameters are tuned and stratified k-fold is applied as well. In both phases, Random Forest classifier performs effectively generating an accuracies of 95.74% and 90.29% in Phase 1 and Phase 2 respectively. Dementia prediction and severity prediction can be extended to provide direct support to members of the assitive care community.