Increasing the rate of occurrence, which is one factor of the significant increase in the amount of big data, is a product brought about by improved sensor technology and lower cost. The frequency of occurrence of data at the site that is a contact point with the real world of information systems is ever increasing. This phenomenon is also called an edge heavy data problem. It is not realistic to transport big data as it is to the cloud, and online computation according to purpose is essential on the spot. In this research, we aim to develop a method combining the excellent points of nonlinear filtering with different characteristics. As a result, we extend the applicability of stream computing technology in machine learning technique.