SELECTION OF PREDICTION MODELS BASED ON POSSIBILITY FUNCTION OF BINARY CONNECTION NUMBERS
Huali Zhou, Huayou Chen · Advances and Applications in Statistics · 2024
Choice of a suitable prediction model is conducive to improve the efficiency of prediction, while the selection of prediction models can be regarded as a problem of ranking predictive performance. On this basis, it is proposed to screen the method of prediction based on the possibility function of binary connection number in this study. First of all, the original time series are transformed by Box-Cox for conformance to the normal distribution of the sequence. Then, the residual components of the decomposed time series are further bootstrapped, and the generated sequences are integrated with the other sub-items. Next, the integrated sequences are transformed by InvBox-Cox to create the training pool. The performance-related index of each prediction model is determined by the training pool. Finally, the possibility function of binary connection number is applied to describe the relative size of the interval number, for comparison of the deterministic and non-deterministic information. The models are sorted to find the one that can be generalized. The method proposed in this study is empirically verified using the PM2.5 data collected in Hefei, China, which shows the rationality and effectiveness of the model. Received: November 6, 2023Accepted: January 8, 2024