SVM Multi-Class Algorithm for Soybean Land Suitability Evaluation

Abstract

Food independence is the ideal of a nation, including Indonesia. However, some agricultural commodities are highly dependent on other countries, such as soybean, one of Indonesia's most popular food ingredients. In recent years, soybean imports have increased in line with the increasing consumption of Indonesian people. The solution that can be offered is to expand soybean farms by considering the land and the weather characteristics. This study aims to evaluate soybean land suitability using a multi-class support vector machine algorithm in the study area of Bogor and Grobogan Regencies. The dataset is divided into two categories, namely explanatory factors consisting of a land slope, soil pH, soil texture, cation exchange capacity, base saturation, depth of mineral soil, rainfall, and temperature, while a target attribute is land suitability classes (very suitable, moderately suitable, marginally suitable, and not suitable). The kernel used to support multi-class properties in SVM includes three, namely polynomial, RBF, and sigmoid. The results show that the RBF kernel model and the 10-fold cross-validation method obtain the best accuracy, 96.91%. In contrast, the model with the combination of the sigmoid kernel and the 5-fold cross-validation method obtained the lowest accuracy, 65.99%.