Incremental Board Learning System Based on Feature Selection

Yi Xu, HuaiDing Qu, Yin Liu, ZhengYue Pan

Abstract: Board learning has similar learning ability and higher learning efficiency as deep learning. The Incremental Board Learning System (IBLS) as an important model of board learning consists of two components: Basic board learning and incremental board learning. However, both components are learned based on all features of the input data, and there are two possible limitations of this method: (1) since features may be redundant, the data set with a large sample size and a large number of features will affect the learning efficiency. (2) Since both components of IBLS are based on all features, the lack of complementarity between features will affect the learning accuracy. For this reason, we propose a new approach: incremental board learning system based on feature selection. First, the feature selection method is used to divide all features into two levels, and the selected features are used as the significant feature layer and the remaining features are used as the normal feature layer. Second, the significant feature layer is used as input for basic board learning to reduce feature redundancy and improve learning efficiency. Third, the normal feature layer is used as the input for incremental board learning to reduce the information loss caused by feature selection and to make full use of the complementarity between the two components of IBLS. Experiments using the proposed method on ten commonly used classification datasets show that it has better learning accuracy and higher classification efficiency than the classic IBLS.