Clustering of Mixed Data: Simultaneous Consideration of Kinematic and Functional Outcome Measures
Kinematic and functional outcome measures obtained from 28 healthy adults (controls) and 27 adults with ankle arthrodesis (patients) were incorporated into a mixed data clustering model. Selected Fourier coefficients were used to parametrize the kinematic data. Principal component analysis (PCA) reduced an original data set of 165 variables to five principal components representing over 75% of the original data variance. A fuzzy clustering algorithm separated the data into 2 clusters, with the dominating principal component (PC) primarily consisting of three functional outcome measures representative of function and pain. Kinematic Fourier coefficients had little effect on the clustering of the data. Approximately half of the patients were clustered with the controls in a high functioning group while the remaining patients were clustered in a low functioning group.