An Efficient Radial Basis Function Neural Network With Hybrid Regularizations for Predicting Library Loan Volume: Generalization and Adaptability Are Improved

Radial basis function neural networks (RBFNNs) have been widely used in data modeling and prediction in recent years. However, an RBFNN does not perform well when it comes to practical problems, such as library loan volume prediction. This happens mainly because the existing RBFNNs overly focus on minimizing training errors, which leads to overfitting and uneven distributions of parameters. In particular, uneven distributions of parameters result in a failure of the RBFNN to cover the dynamic characteristics of the library loan volume. To solve this problem, this article proposes an efficient RBFNN with hybrid regularizations (RBF-HR) for predicting the library loan volume. In the proposed RBF-HR, we design an efficient dynamic repulsion regularization (DRR) of center and adaptive regularization of width to improve the generalization and adaptability for practical problems. We also verify the effectiveness of the proposed RBF-HR in predicting the library loan volume.