A research of intelligent parameters searching in small data sets

Increasingly competitive in a global economy, the lifecycle of product become shorter and shorter. How to shorten the time during research and development period, especially in the early stage in the industrial lifecycle is now an important issue. Unfortunately, lack of sufficient data always is a problem while acquiring knowledge in early stage. Therefore, this paper focuses on small data sets and further provides a systematic way for parameters searching. Our methodology is effectively selecting experimental parameter settings for redefining the boundary of parameter settings iteratively. There are four stages in our methodology: virtual sample generation, classification, selection and performance testing. In this paper, we design two experiments for verification four different selection mechanisms (RS, SVS, LVS, GVS). Furthermore, LVS and GVS mechanism will be discussed in the convergence experiment.