Deep Learning-based query-count forecasting using farmers’ helpline data

Abstract

Agriculture plays a vital role in the global economy, provides the primary source of food, livelihood, and employment to the populations.Therefore, we are in a continuous quest to develop innovative approaches with the intention to boost agricultural productivity. For planning agricultural-policies, organizing farmers' training, gaining agricultural product-based market insights, and executing strategic marketing actions, government officials, policymakers, and private organizations need to gain awareness regarding farmers' problems. However, today, there is a need of such a robust system that can be used to collect and analyze Spatio-temporal information regarding the problems faced by farmers on a large scale. This article describes a new approach that uses high-end artificial intelligence-based techniques to forecast the trends in agricultural problems using farmers' helpline data. The dataset utilized in this work is accumulated from the ``Kisan Call Center", a farmers' helpline center administered by the Ministry of Agriculture, Government of India, from March 2013 to March 2020. Moreover, we take data corresponding to the top-five rice-producing states of India (Uttar Pradesh, Punjab, Bihar, West Bengal, and Andhra Pradesh) as case studies to inspect the performances of the proposed framework with four different forecasting periods (1, 7, 15, and 30-days forecasting). Later, we compare the forecasting potential of four different Machine Learning and Deep Learning-based forecasting techniques, i.e., Support Vector Regression, Multi-layer Perceptron, Long Short-Term Memory Networks, and Gated Recurrent Units using three different performance measures (Mean Squared Error, Mean Absolute Error, and Correlation Coefficient). From the experimental results, we found that the proposed framework is useful for forecasting trends in farmers' problems, furthermore, we identify various other potential applications of the presented work. Finally, we conclude with some possible future developments in the proposed approach.