Hybrid linear time series approach for long term forecasting of crop yield

Long term forecasting of crop production is required to establish long term vision, say by 2025, to meet growing demand of population at that point of time. Existing univariate linear time series ARIMA approach is valid for short term forecast only. In this paper, a technique for long term yield for...

Full description

Saved in:
Bibliographic Details
Published in:The Indian journal of agricultural sciences 2018-08, Vol.88 (8), p.1275-1279
Main Authors: ALAM, WASI, SINHA, KANCHAN, KUMAR, RAJEEV RANJAN, RAY, MRINMOY, RATHOD, SANTOSHA, SINGH, K N, ARYA, PRAWIN
Format: Article
Language:English
Subjects:
ARIMA
Artificial neural network
Hybrid ARIMA
Long term yield forecast
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Long term forecasting of crop production is required to establish long term vision, say by 2025, to meet growing demand of population at that point of time. Existing univariate linear time series ARIMA approach is valid for short term forecast only. In this paper, a technique for long term yield forecast has been proposed. Initially, we have tried to improve short term forecast of yield by using hybrid ARIMA through ANN approach. The forecast values of yield through hybrid approach was considered as baseline data for long term forecast of yield. Time series data on rice yield was considered for Aligarh district of Uttar Pradesh for the study. Through ARIMA (2,1,0), we got short term forecast of yield by 2020 and the residuals obtained by 2013 were used to model and forecast through ANN approach. For the residuals, 05:04s:1l (05 time delay and 04 hidden nodes) model was identified as suitable one as it has minimum values of mean absolute percentage error (MAPE) for training and testing sets. Using 05:04s:1l model, residuals were forecasted by 2020, forecast values of yield obtained through ARIMA (2,1,0) were corrected by forecasted residuals and eventually get forecast of yield through hybrid approach. The estimated MAPE for ARIMA (2,1,0) and hybrid approach were 17.677% and 4.65%, respectively. Significant reduction in MAPE through hybrid approach indicates it’s much better performance as compared to ARIMA alone. Using hybrid approach, we got forecast of yield by 2020 and considering this forecasted yield as baseline data, we got forecast by 2025 through the proposed approach.
ISSN:0019-5022
2394-3319
DOI:10.56093/ijas.v88i8.82573