Title: Hybridising neural network and pattern matching under dynamic time warping for time series prediction
Authors: Thanh Son Nguyen
Addresses: Faculty of Information Technology, Ho Chi Minh City University of Technology and Education, 01 Vo Van Ngan, Linh Chieu Ward, Thu Duc District, Ho Chi Minh City, Vietnam
Abstract: Pattern matching-based forecasting models are attractive due to their simplicity and the ability to predict complex nonlinear behaviours. Euclidean measure is the most commonly used metric for pattern matching in time series. However, its weakness is that it is sensitive to distortion in time axis; so, this can influence on forecasting results. The dynamic time warping (DTW) measure is introduced as a solution to the weakness of Euclidean distance metric. In addition, artificial neural networks (ANNs) have been widely used in the time series forecasting. They have been used to capture the complex relationships with a variety of patterns. In this work, we propose an improved hybrid method which is an affine combination of neural network model and DTW-based pattern matching model for time series prediction. This method can take full advantage of the individual strengths of the two models to create a more effective approach for time series prediction. Experimental results show that our proposed method outperforms neural network model and DTW-based pattern matching method used separately in time series prediction.
Keywords: time series; pattern matching; artificial neural network; ANN; time series prediction; dynamic time warping; DTW; k-nearest neighbour.
International Journal of Business Intelligence and Data Mining, 2020 Vol.17 No.1, pp.54 - 75
Received: 21 Aug 2017
Accepted: 05 Dec 2017
Published online: 05 Apr 2020 *