STUDY ON MACHINE LEARNING TECHNIQUES BASED HYBRID MODEL FOR FORECASTING IN AGRICULTURE
Loading...
Date
2019
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
ICAR-Indian Agricultural Statistics Research Institute ICAR-Indian Agricultural Research Institute New Delhi
Abstract
Agricultural datasets are mostly nonlinear, nonstationary and leptokurtic in nature. These properties
of dataset pose a variety of problems in forecasting. Precise forecasting helps both farming
community and policy makers to undertake informed decisions. Literature suggests that each of the
forecasting models has their own limitations. A single forecasting model is not able to handle
problems like nonstationary and nonlinearity simultaneously. Accordingly, the present study
proposes three different hybrid models i.e. empirical mode decomposition based support vector
regression (EMD-SVR), time-delay neural network with error correction term (TDNN-ECT) and
multivariate adaptive regression splines based artificial neural network (MARS-ANN) models. The
novelty of these models lies in the fact that they can handle both nonstationary and nonlinear
features of dataset simultaneously. In EMD-SVR model, the nonstationary and nonlinear dataset is
decomposed into different intrinsic mode functions and final residue through EMD method. Then
the decomposed components are forecasted using SVR model and finally, all forecasted values are
summed up to produce the final forecast. In the second model, TDNN-ECT uses the error correction
term from the two co-integrated series as auxiliary variable. The auxiliary information in the form
of ECT improves the forecasting accuracy. Further, selection of important input variables is a
crucial step in determining the accuracy of any forecasting model. Hence, MARS-ANN hybrid
model was developed in which the MARS algorithms was employed to extract important factors
determining crop yield and the extracted factors were used for yield prediction using ANN
methodology. The performance of proposed hybrid models is evaluated with individual forecasting
models using three different agricultural datasets. The performance measures like RMSE, MAD,
MAPE and ME are used to evaluate the model. The results indicated that the performance of the
proposed hybrid models are substantially superior as compared to the individual forecasting model.
Key words: Co-integration, Nonlinearity, Nonstationary, EMD, SVR, TDNN, and MARS.
Description
T-10279
Keywords
null