This study aims to develop forecasting method based on sequential version of Relevance Vector Machine (RVM). The main idea is to perform the general optimization of the weights and hyperparameters using the current relevance vectors and newly arriving...
This study aims to develop forecasting method based on sequential version of Relevance Vector Machine (RVM). The main idea is to perform the general optimization of the weights and hyperparameters using the current relevance vectors and newly arriving data. By doing so, sequential RVM are trained with more currently arrived data. It extends RVM algorithm to real-time and non-stationary learning processes. The accuracy and the learning time of the proposed algorithm are better than the Support Vector Machine (SVM), original RVM, and Gaussian Processes (GP).