Development of Prediction System for Crude Palm Oil (CPO) Production with Time Series Data Mining Approach

Abstract

Palm oil is a plantation commodity that snowballs when compared to other plantation crops such as coffee or cocoa. The Indonesian palm oil industry has a comparative advantage in the form of a large area of land and the lowest production cost of Crude Palm Oil (CPO) in the world. Indonesia's palm oil production in August 2019 recorded an increase of 14% over the same period in 2018. However, the amount of Indonesia's CPO production can still be optimized and increased. The amount of CPO production is very dependent on several factors, such as weather conditions, land area, and the number of Fresh Fruit Bunches (FFB). To help the Palm Oil Mill (POM), this study compares three data mining algorithms to predict the amount of CPO production based on the number of FFBs. The algorithms being compared are multilayer perceptron (MLP), support vector regression (SVR), and linear regression (LR). Based on test results using test data from a palm oil company in Indonesia, the SVR algorithm can provide higher accuracy than the other two algorithms. The SVR gets a PTA value of 0.694, MSE of 955.002, MAPE of 55.169, and MAD of 22.227. Then, we developed a prototype that applied the SVR algorithm to predict the amount of CPO production. The SQA test results on the prototype resulted in 80.225 software quality in the good category.