Comparing Unsupervised Machine Learning Algorithms to Cluster the Responses to the Canadian Community Health Survey

• In Event: Stressors in the Lives of Students

Abstract

Mental illness is a pressing medical challenge, thus, identifying gaps in mental-healthcare support-seeking is essential to public health. We employed unsupervised machine learning to reveal patterns and gaps of mental-healthcare support utilization in the latest CCHS-MH sample of n=24,788 Canadians. Three clustering methods (K-Means, Hierarchical Agglomerative, and Fuzzy C-Means) were compared. Fuzzy C-Means yielded the best-fit model for the survey data. Exploratory Factor Analysis revealed an optimal four-factor model (Mixed Support, No Support, Social Support, and Family Doctor). Findings suggest that rural Canadians are not utilizing social support as often as their metropolitan counterparts, women rely on social support but under-utilize family doctors, and middle-aged individuals overwhelmingly seek no support. Implications for increased positive engagement with these groups are discussed.

Authors

• Chang Lu, Shanghai Jiao Tong University

• Jordan Southcott, University of Alberta

• Maria Cutumisu, University of Alberta