複数自己組織化マップを用いた欠損データの推定

概要

論文の詳細を見る
In this paper, we propose a new method that uses multiple SOMs for estimating values lacking in data analysis. Recently, development of information technology grows the importance of data analysis. In actual data, however, some values will be sometimes missing, and then dealing with such insufficient data has become one of the important subjects in data analysis. Estimating and completing the empty values are required to applying various data analysis techniques. Such an estimation method is also applicable to data prediction problems. In the former methods that use SOM, many empty values would have caused the lack of data for learning process. Our system can achieve effective learning by using multiple SOMs even for data that includes many missing values. Moreover, the system is still available for nonlinear data because of using SOMs. We performed some numerical simulation using the proposed and other methods. By the simulation results, we showed the advantages of our method over some traditional techniques including a technique that uses single SOM.
一般社団法人日本機械学会の論文