A Time Series Forecasting Model Selection Framework using CNN and Data Augmentation for Small Sample Data

Abstract

The key to the accuracy of time series forecasting is to find the most appropriate forecasting method. Therefore, the forecasting model selection of time series has become a new research hotspot in the data analysis field. However, most of the existing forecasting model selection methods reduce the forecasting efficiency for relying on subjective manual selection of features. In this paper, an automatic time series feature extraction framework is proposed for forecasting model selection based on the idea of meta learning. Inspired by computer vision, we transform one -dimensional time series into two-dimensional images, and use convolution neural network (CNN) to train and classify time series images (model selection). Moreover, in order to deal with the over fitting problem caused by small sample datasets, the sliding window data augmentation method is used to improve the accuracy of small datasets model selection. A large-scale empirical study on M3 datasets shows that the proposed framework has better model selection accuracy and smaller forecasting error(MAPE) than Support vector machine(SVM) and traditional time series image algorithms. Moreover, the classification rate(model selection accuracy) of the proposed algorithm are increased by 6.5% and 4.4% compare with the traditional time series image method and Support vector machine respectively in average. c