Kullback chi square and Gustafson Kessel probabilistic neural network based soil fertility prediction

Abstract

AbstractAgriculture plays a significant role in any country's wealth creation, fulfilling food security needs and employment generation thereby contributing immensely to the country's growth and GDP. However, certain characteristics such as climate and environmental changes have extensively become menacing factors in the agriculture output. As soil is an influential specification influencing the prediction of crop yield, soil analysis becomes imperative and can assist farmers in preliminary adaptations towards better crop planning thereby facilitating higher yields. Machine learning algorithms have materialized in soil fertility prediction as an encouraging method for enhancing production. However, the spread and usage of this method are still limited by the lack of clear applicability due to the uncertainties involved and therefore resulting in false predictions by farmers. In this work, we take into account and address both the problems by incorporating uncertainty quantification utilizing the fishers ratio preprocessing model and Kullback divergent chi‐square feature selection for soil fertility prediction. Next, Gustafson‐Kessel probabilistic neural network classification uses the soil fertility predictive model to produce the probability distribution as output and the different types of soil fertility level instead of a single value. We analyze and prove that the new method not only provides uncertainty quantification but also minimizes the processing time and false positive rate with high accuracy than the existing soil fertility prediction methods.