Comparison of K-Nearest Neighbor and Support Vector Regression for Predicting Oil Palm Yield

With the high demand for oil palm production, implementations of Machine Learning (ML) technologies to provide accurate predictions and recommendations to assist oil palm plantation management tasks have become beneficial, such as in predicting annual oil palm. However, different geographical and meteorological conditions may result in different scales of influence for each variable. In this research, K-Nearest Neighbors (KNN) and Support Vector Regression (SVR) were used in predicting oil palm yield based on data collected in Riau, Indonesia. Pearson’s correlation coefficient was also calculated in selecting the input features for the models, whereas normalization and standardization were used in scaling the data. By setting the minimum absolute correlation threshold to 0.1 and using standardization, both models managed to obtain more than 0.81 2 , with SVR obtaining the overall best performance with 0.8709 2 , 1.372 MAE, and 1.8025 RMSE without hyperparameter fine-tuning. It was also discovered that the oil palm yield in the previous year is the variable with the most influence in estimating oil palm yield in the current year, followed by the number of plants and soil types.

Authors:
Bens Pardamean, Teddy Suparyanto, Gokma Sahat Tua Sinaga, Gregorius Natanael Elwirehardja, Erick Firmansyah, Candra Ginting, Hangger Gahara Mawandha, and Dian Pratama Putra

International Conference on Innovative Technologies in Intelligent Systems and Industrial Applications, CITISIA 2022

Read Full Article