Climate Based Crop Recommendation System

Abstract

Agriculture yield depends on the climate and geological factors. Choosing the newlineright crop at the right time is the most important factor in obtaining a higher yield. newlineTherefore, any information associated with climatic factors will ensure farmer s newlineforeordained farming. The proposed work puts forward a recommendation engine newlinecomprising of four models, viz., weather forecasting, crop classification, crop type newlineclassification, and district classification. The weather forecasting model is built newlineto predict weather for a long range of twelve months. Different attributes that newlineinclude cloud cover, diurnal temperature range, maximum and minimum temperatures, newlinevapour pressure, potential evapotranspiration, precipitation, wet day newlinefrequency, relative humidity, reference crop evapotranspiration, and ground frost newlinefrequency are used in the development of weather forecast models. A confidence newlineinterval of 95% is computed to determine a target average Mean Absolute Error newline(MAE) for the weather forecast model. Deep learning algorithms like Recurrent newlineNeural Network (RNN), Gated Recurrrent Unit (GRU), and Long Short Term newlineMemory (LSTM) are used to build three different forecast model and the errors of newlineall these models are compared and found to be lower than the target error, with newlinethe LSTM model being the closest. Hence, to enhance the LSTM model and attain newlinethe target error, moving averages of the forecasting attributes are included as newlineadditional features. Thus, a new feature-engineered moving average based LSTM newlinemodel is proposed. The proposed weather forecast model is compared with statistical newlineweather forecasting models like Simple Exponential Smoothing, Seasonal newlineAuto-Regressive Integrated Moving Average with eXogenous factors (SARIMAX), newlineHolt s method, and Holt Winter Exponential Smoothing. For building crops, crop newlinetypes, and district classification models, various attributes of 105 crops in the state newlineof Tamil Nadu are collected from different sources and curated as a comprehensive newlinedataset.