Classification models using the conversion of shape landmark data

Dryden, Mardia (2016) define shape analysis as an analysis that measures and describes the shapes of objects represented by landmarks in geometric space and compares them. The shape point data used for shape analysis has the form of an array, which is a form that is difficult to apply to the shape classification model. This study presents two methods for applying shape point data to the classification model. The first method is to apply the Generalized Prospects Analysis (GPA) fit and then obtain Riemannian distance, centroid size, and shape PC scores to select them as variables, and the second method is to vectorize the configuration matrix of each sample to use all coordinate values as variables. The data obtained by these two methods are put into three classification models: Random Forest, Logistic Regression Analysis, and SVM to compare the misclassification rates for each data. To this end, the method above was applied to the mice data from Dryden, Mardia (2016), and the results were examined. The comparison revealed that the mice data of three clusters, which is a 2-dimensional data set with a small number of shape points, showed that the rotation parameter had a significant impact on cluster determination. Additionally, vectorization of the original data or centered data, which had not undergone the fitting process, yielded better results.