Research on a dynamic full Bayesian classifier for time-series data with insufficient information

Small sample time-series data with insufficient information are ubiquitous. It is challenging to improve the classification reliability of small sample time-series data. At present, the dynamic classifications for small sample time-series data still lack a tailored method. To address this, we first setup the architecture of dynamic Bayesian derivative classifiers, and then establish a dynamic full Bayesian classifier for small sample time-series data. The joint density of attributes is estimated by using multivariate Gaussian kernel function with smoothing parameters. The dynamic full Bayesian classifier is optimized by splitting the smooth parameters into intervals, optimizing the parameters by constructing a smoothing parameter configuration tree (or forest), then selecting and averaging the classifiers. The dynamic full Bayesian classifier is applied to forecast turning points. Experimental results show that the resultant classifier developed in this paper is more accurate when compared with other nine commonly used classifiers.