Classification of Salak Sidempuan Productivity Using Machine Learning Based on Soil Fertility and Rainfall Data in North Sumatra, Indonesia

Abstract

Land productivity assessment is important for understanding variation in Salak Sidempuan production under different soil and climatic conditions. This study aimed to classify Salak Sidempuan productivty using soil fertility and annual rainfall data and to compare the performance of five machine learning algorithms. Data from 500 Salak Sidempuan tree observations were collected from Marancar, West Angkola, and South Angkola, South Tapanuli, Indonesia. The predictor variables consisted of soil pH, organic carbon, nitrogen, phosphorus, potassium, and annual rainfall, while productivity was divided into low, moderate, and high classes. Decision Tree, Random Forest, Support Vector Machine, XGBoost, and LightGBM were optimized using 5-fold cross-validation on the training dataset and evaluated using an independent testing dataset. Among the evaluated models, Random Forest achieved the highest performance, with an accuracy of 66.40%, precision of 67.01%, recall of 66.47%, and F1-score of 66.46%. Feature importance analysis showed that annual rainfall had the highest relative contribution (24.8%), followed by C-organic (22.1%) and nitrogen (17.6%), while potassium, phosphorus, and soil pH contributed 13.9%, 11.8%, and 9.8%, respectively. These findings indicate that soil fertility and rainfall variables provide useful information for distinguishing Salak Sidempuan productivity classes. However, the moderate classification performance indicates that the selected variables do not fully explain productivity variation. The results therefore provide an initial data-driven assessment of Salak Sidempuan productivity based on the environmental variables included in the study.