Application of Machine Learning Algorithms for Soil Erosion Susceptibility Estimation in Gumani River Basin, Eastern India

Abstract

Abstract The primary goal of this study is to determine soil erosion risk susceptibility and to suggest the most appropriate techniques for soil erosion susceptibility for Gumani river basin. This research examined the point-specific values of several components derived from randomization. The susceptibility of the projected frameworks, namely the Artificial Neural Network and Support Vector Machine, was investigated with the help of the most significant causative variables and the corresponding field records. The area underneath the Receiver Operating Characteristics curve shows accuracy for ANN and SVM accordingly. For assessing susceptibility, the ANN (AUC = 0.932) and SVM (AUC = 0.915) were used for training points, whereas ANN (AUC = 0.906) and SVM (AUC = 0.882) were for validation points. The ANN model is very efficient in simulating the erosional and non-erosional regions more accurately than SVM. The outcome of ANN predicted that 19.14% area is very high, extended in the entire western parts and some parts of the southern part, 14.96% is high, extended in the entire western part, 16.01% is moderate, extended in the western and eastern parts, 18.54% low, and 31.35% very low extended in the middle and eastern parts, susceptible for soil erosion whereas the outcome of SVM predicted that 15.45% area is very high, 18.82% high, 22.05% moderate, 22.94% low, 20.74% very low, susceptible for soil erosion. The land degradation phase is not a unidirectional process. Therefore, multidimensional effects from conditioning factors must be calculated accurately by considering the maximum possible variables and choosing optimum models for particular areas. These attempts will help policymakers implement proper methods to check soil erosion in the Gumani river basin.