Determining soil fertility using principal componen regression analysis of oil palm plantation in West Sulawesi, Indonesia

Abstract

Palm oil is one of the most popular vegetable oils in the world besides soybean oil. Nowadays, more than 80% of the world's palm oil supply comes from Indonesia and Malaysia. As the largest supplier, Indonesia must be improved and maintained its productivity with good management, such as maintaining soil fertilization to produce the best oil palm. This activity requires analysis of the soil chemical factors and soil conditions (fertile or not) from oil palm plantations, which involves a large amount of data. Principal component regression is an appropriate statistical method to solve the problem. In this method, the coefficient of the regression model is the coefficient of the principal components (PCs). Dimension reduction is performed on the predictor variables that have a high correlation so that the PCs have an insignificant correlation. The data are obtained by measuring soil samples around the trees from oil palm plantation in West Sulawesi, Indonesia. The data consist of sixteen variables and thirty-six observations (0-20 cm) for each variable. There are three PCs that become predictors of the regression model with information absorption rates reaching 78%, i.e., some macronutrients (Potassium oxide, Potassium, Calcium, and Cation exchange capacity), soil acidity and organic properties (Carbon and Nitrogen). Furthermore, the accuracy of the estimation value in this logistic regression model reaches 90% by using stepwise backward method.