Analysis and Prediction of Cotton Yield with Fertilizer Recommendation Using Gradient Boost Algorithm

Abstract

The majority of every nation’s economy depends on agriculture. However, unpredictable changes in climatic conditions have affected the cultivation of crops. There is a declining interest for individuals to remain in this field since farmers are not encouraged to use technology for processes. This happens primarily because there is a lack of involvement of information technology in the farming sector. This obstacle can be overcome by applying several techniques of programming to help farmers estimate the yield of their produce and motivate them to remain in this field. Crop yield prediction includes anticipating yield of the crop from available historical information like climatic, acreage parameters and soil parameters. Machine learning is one technique that can be used to predict the crop yield. Using parameters that are associated with the environment, agriculture and farming, machine learning models can predict the crop yield with good accuracy. The research previously done in this sector has proven to be instrumental in forming the basis of this model. The findings have shown that parameters such as nutrient levels, acreage and historical data have a greater impact on the overall yield in addition to the climatic conditions of the place being observed. In this work, a model has been proposed which helps cultivators predict the yield of the crop even before cultivating directly onto the agricultural lands. The yield prediction based on location, acreage and fertilizer data for several regions in the United States has been experimented using various algorithms. The parameters tested and the features used introduce a unique insight into the relationship between nitrogen, phosphorous and potash (N-P-K) used (acreage covered and quantity) and the lint yield of cotton. A point to note here is that the model has taken into account the historical observations (1975–2017) in order to recommend fertilizers. This has been made possible using the multi-variate regression algorithm. Additionally, the yield prediction results have shown that the gradient boosting regressor proved to be the best performer, returning a standalone accuracy score of 90.73% and 80.72% on being validated using the fivefold cross validation method. The proposed model aims to design, develop and implement the training model by using different input data. The machine will be able to learn the features and extract the crop yield from the data by using data mining and data science techniques.