A Novel Time Series Forecasting Approach Considering Data Characteristics

Abstract

A novel time series forecasting approach with consideration of inner knowledge hidden in data, in terms of data characteristics, is proposed. In the proposed methodology, the main data characteristics hidden in the observed time series data are first explored; and according to the data characteristics, suitable forecasting models are formulated to improve prediction performance. For illustration, the proposed methodology is used to predict Chinese total social consumption and total energy consumption. The empirical results show the forecasting model considering data characteristics outperforms other popular forecasting models ignoring data characteristics, which further implies that data characteristics exploration is an important and necessary step in forecasting and the proposed methodology can be used as a promising approach for time series forecasting.

An elm-based classification algorithm with optimal cutoff selection for credit risk assessment

“…The hybrid approaches coupling AI techniques and other traditional methods have been shown even more powerful, such as some ANN-based forms combining the ANN with the models of DA [28][29][30], multivariant DA (MDA), Iterative Dichotomizer 3 (ID3) method [31], and clustering analysis [32]. Since the AI techniques and their hybrids have repeatedly been shown much more powerful than other traditional models [33,34], this paper tends to conduct the credit risk assessment study based on the AI tools and hybrid concept.…”

“…Though the AI models have been shown powerful in credit risk assessment compared with traditional models, they have their own limitations, e.g., time-wasting and local minima [34]. To address these problems, extreme learning machine (ELM), a special case of single hidden layer feedforward networks (SLFNs), was currently proposed by Huang et al [35].…”