Predictive Analysis to Improve Crop Yield using a Neural Network Model

Agriculture has been the sector of paramount importance as it feeds the country population along with contributing to the GDP. Crop yield varies with a combination of factors including soil properties, climate, elevation and irrigation technique. Technological developments have fallen short in estimating the yield based on this joint dependence of the said factors. Hence, in this project a data-driven model that learns by historic soil as well as rainfall data to analyse and predict crop yield over seasons in several districts, has been developed. For this study, a particular crop, Rice is considered. The designed hybrid neural network model identifies optimal combinations of soil parameters and blends it with the rainfall pattern in a selected region to evolve the expectable crop yield. The backbone for the predictive analysis model with respect to the rainfall is based on the Time-Series approach in Supervised Learning. The technology used for the final prediction of the crop yield is again a branch of Machine Learning, known as Recurrent Neural Networks. With two inter-communicating data-driven models working at the backend, the final predictions obtained were successful in depicting the interdependence between soil parameters for yield and weather attributes.