Development of probabilistic transition models using self-organizing of a large-scale histrical data on space-time dimension

点击翻译按钮获取中文摘要

When we analyze a large-scale historical data on space-time dimension, there is the difficulty with the variety of the history. In this study, we proposed methodologies to develop probabilistic models of the transitions in histories, using their topological structure and the structure of the subsequence of history. We develop flexible self-organization map (FSOM), which is a SOM whose map range is automatically adjusted by the topological structure of history. Transitions in histories are grasped by the structure derived from the FSOM and so on. We verified our methodologies with the real world data, for example, service use history in merchandising or the finance business, health care service histories such as the receipt, and movement histories in city space.