Fusion of statistical and machine learning approaches for time series prediction using earth observation data. (2017)

• Record Type: Journal Article
• Authors: Agrawal, K.P.; Garg, Sanjay; Sharma, Shashikant; Patel, Pinkal; Bhatnagar, Ayush
• Is Part Of: International journal of computational science and engineering. Volume 14:Number 3(2017)
• Page Start: 255
• Page End: 266
• Publication Date: 2017
• Subjects: prediction -- time series -- auto regressive integrated moving average -- ARIMA -- support vector regression -- SVR -- scalability; Computer science -- Mathematics -- Periodicals; Computer simulation -- Mathematical aspects -- Periodicals; Computational intelligence -- Periodicals; 004.015105
• Languages: English
• ISSNs: 1742-7185

Abstract

This paper focuses on fusion of statistical and machine learning models for improving the accuracy of time series prediction. Statistical model like integration of auto regressive (AR) and moving average (MA) is capable to handle non-stationary time series but it can deal with only single time series. The machine learning approach [i.e. support vector regression (SVR)] can handle dependency among different time series along with nonlinear separable domains, however it cannot incorporate the past behaviour of time-series. This led us to hybridise auto regressive integrated moving average (ARIMA) with SVR model, where focus has been given on minimisation of forecast error using residuals. Issue related to scalability has been handled by taking suitable representative samples from each sub-areas which helped in reducing number of models and drastic reduction in time for tuning different parameters. Results obtained show that the performance of proposed hybrid model is better than individual models.