Abstract
In this paper, we propose Principal Component Analysis (PCA) for modal interval-valued data. Professor Diday proposed Symbolic Data Analysis (SDA) which aggregate large datasets without loss of essential information. The aggregated data is called symbolic data. We focus on PCA in the SDA framework. Most studies on PCA in SDA assume interval-valued data. However, if we use interval-valued data as an aggregation, the variation of the original data is often ignored. The variation might have essential information and should be utilized. We introduce modal interval-valued data represented by histograms to PCA. We evaluate and compare the results of the proposed method and those of conventional methods through simulations. In addition, we apply our method to real data about TV drama and show its availability.
