부산대학교 도서관

Longitudinal data measured repeatedly over multiple time points for the same object can be analyzed through the generalized estimating equation (GEE) or the generalized linear mixed model (GLMM), and so on. However, the study on the visualization techniques is somewhat lacking and has some limitations. They tend to be very simple level of visualization and can only be applied to contingency tables of a particular size. And, even if visualization techniques applicable to all size contingency table, it is difficult to identify and interpret changes over time at a glance, because it becomes complicated when the number of variables or categories increases. In addition, the distance between the row and column coordinates is not geometrically meaningful and only the direction in which the two coordinates are placed can explain the relation between the row and column categories. Also, many studies use only variables information alone without using objects information.Therefore, in this thesis, we develope a visualization algorithm of multidimensional longitudinal data based on the projection method using the indicator matrix. In this case, the distance between two coordinates can be used to identify the relative relation between two categories. Also, by using the indicator matrix with objects information, so that the relations between categories and objects, and the clustering of objects can also be possible. Through this algorithm, we will propose visualization of longitudinal data analysis plot (VLDA plot). Furthermore, we will propose an algorithm for supplementary objects and variables that can often occur in longitudinal data.