Time Series Prediction using ARIMA and DBNs with MODWT
Keun-Tae Park, Jun‐Geol Baek · Journal of Korean Institute of Industrial Engineers · 2017
Times series data is closely related with real-world more than other data. Time Series Prediction is one of the most important subjects that is useful in real-world problems. There are already many time series analysis methods. This study try to overcome the limitations of one of the famous time series analysis methods, ARIMA. ARIMA has limitations that are weakness in short term prediction and absence of nonlinear pattern analysis. This study use MODWT (Maximum Overlap Discrete Wavelet Transform) for preprocessing the time series data, and predict the data with ARIMA (Auto-Regressive Integrated Moving Average) and DBNs (Deep Belief Networks) which is usually used for analyzing nonlinear data. Real case datasets are used to compare the performances with original ARIMA and existing prediction methods. The results from the experiments demonstrate the usefulness and possibilities in various time series fields and superiority with improved accuracy.