The aim of this research is forecasting crude oil prices using Support Vector Regression (SVR). Algorithm to determine the optimal parameters in the model using the SVR is a grid search algorithm. This algorithm divides the range of parameters to be optimized into the grid and across all points to get the optimal parameters. In its application the grid search algorithm should be guided by a number of performance metrics, usually measured by cross-validation on the training data. Therefore, it is advisable to try some variations pair hyperplane parameters on SVR. Based on analysis calculation of accuracy and the prediction error using the training data generating R2 99.10868 % while the value of MAPE by 1.789873 % . The data testing generates R2 96.1639% while the value of MAPE by 1.942517 %. This indicates to the data of testing using a linear kernel or accuracy of prediction accuracy results are quite large. Best model using the SVR has been formed can be used as a predictive model of crude oil prices. The results obtained showed crude oil prices from period 1 up to 10 experiencing decline.

Keywords: Crude Oil Prices, SVR, Kernel, Grid Search, Cross Validation

  1. Introduction

Final energy consumption in Indonesia for the period 2000 – 2012 increased by an average of 2.9% per year. The most dominant type of energy is petroleum products which include aviation fuel, avgas, gasoline, kerosene, diesel oil, and fuel oil. These types of fuel consumed mostly by the transport sector.  Today, most of the fuel prices are still subsidized. Fuel subsidies in 2013 have reached 199 trillion rupiahs. The government is also still subsidizing electricity for a particular type of users. Total electricity subsidies in 2013 reached 100 trillion rupiahs. The energy subsidy (fuel and electricity) has been increasing steadily. Energy subsidies in 2011 amounted to 195.3 trillion rupiahs and increased to 268 trillion rupiahs in 2013. Total spending on energy subsidies is always greater than the allocated budget and it often causes problems by the end of each fiscal year. Caraka and Yasin (2014) introduced the government has issued a number of policies to reduce petroleum fuel usage. Crude oil price is based on January 2016 data with 22.48 $/barrel (current price) and it assumed to be rising linearly to 40 $/barrel in the end of 2016. Oil production continues to decline while the demand for energy continues to grow which led to the increase in import of crude oil and petroleum products. This was shown by the deficit 3,5 billion Dollar at oil account in the second quarter which increased from 2,1 billion Dollar deficit in the first quarter of 2014 financial year. On the other hand, fuel subsidy is relatively high, due to increased domestic consumption, the increase in international oil prices and the decline in the exchange rate against the dollar and other foreign currencies. It is estimated that fuel subsidies until the end of 2014 will exceed the budget allocation in 2014. Since the publication of the 2015 edition of the WOO in November last year, the most obvious market development has been the oil price collapse. While the average price of the OPEC Reference Basket (ORB) during the first half of 2014 was over $100/barrel, it dropped to less than $60/b in December 2014 and has averaged close to $53/b in the first nine months of 2015. This new oil price environment has had an impact on both demand and supply prospects in the short- and medium-term, and some lasting effects can be expected in the long-term. Crude oil prices are expected to remain low as supply continues to outpace demand in 2016 and more crude oil is placed into storage. EIA estimates that global oil inventories increased by 1.9 million b/d in 2015, marking the second consecutive year of inventory builds. Inventories are forecast to rise by an additional 0.7 million b/d in 2016 before the global oil market becomes relatively balanced in 2017. The first forecasted draw on global oil inventories is expected in the third quarter of 2017, marking the end of 14 consecutive quarters of inventory builds. In the time domain, the long memory is indicated by the fact that the oil prices eventually exhibit strong positive dependence between distant observations. A shock to the series persists for a long time span even though it eventually dissipates. In the frequency domain, the long memory is indicated by the fact that the spectral density becomes unbounded as the frequency approaches zero.

  • Basic Concept of SVR

Santosa (2007) explained that Support Vector Machines (SVM) is a technique to make predictions, both in the case of classification and regression. SVM with linear classifier has a basic principle which is the case classification linearly separable, but SVM has been developed in order to work on a non-linear problem by incorporating the concept of the kernel in high-dimensional space (Gunn, 1998). By using the concept of ε-insensitive loss function, which was introduced by Vapnik, SVM can be generalized to approach the function or regression Support Vector Regression (SVR) is the application of SVM for regression case. In the case of regression output as real numbers or continuous. SVR is a method that can overcome the overfitting, so it will produce a good performance. Suppose there are  training data ,  where  an input vector  and scalar output  and  is the number of training data. With the SVR, we want to assign a function  which has the greatest deviation  from the actual target , for all of the training data. If the value of  is equal to 0 then obtained a good regression equation.


This research is supported by Directorate of Research and Community Services, The Ministry of Research, Technology, and Higher Education.


  • Abe, S. 2005. Support Vector Machine for Pattern Classification. Springer – Verlag. London Limited.
  • Caraka,R.E.,Yasin,H. 2014. Prediksi Produksi Gas Bumi dengan General Regression Neural Network (GRNN), in Proc.National Statistics Conference (SNS IV). ISSN : 2087-2590, No 270-277
  • Gunn, S. 1998. Support Vector Machines for Classification and Regression. Technical Report, ISIS.
  • Hamel, L. 2009. Knowledge Discovery With Support Vector Machine. United State Of America :A John Wiley &Sons, INC., Publication.
  • Santosa, 2007. Data Mining Teknik Pemanfaatan Data untuk Keperluan Bisnis. Yogyakarta:Graha Ilmu.
  • Scholkopf B., and Smola A. 2003. A Tutorial on Support Vector Regression.
  • Yao, Y, et al. 2014. An Improved Grid Search Algorithm and Its Application in PCA and SVM Based Face Recognition, Journal of Computational Information Sistems. Vol. 10, No. 3:1219–1229.
  • H., Prahutama.A., and Utami,T.W. 2014. Prediksi Harga Saham Menggunakan Support Vector Regresion dengan Algoritmaa Grid Search, Statistika. Universitas Diponegoro. Media Statistika. Vol 7, No 1:29-35

For further information please refer to GJPAM



Leave a Reply

Your email address will not be published. Required fields are marked *