Application of support vector machines to the forecast of direction of exchange rate in Nigeria

Alabi, Nurudeen Olawale and Are, Stephen Olusegun (2018) Application of support vector machines to the forecast of direction of exchange rate in Nigeria. In: Federal Polytechnic, Ilaro International Conference. Federal Polytechnic, Ilaro, November 8, 2019, The Federal Polytechnic, Ilaro.

Text Conference 3_Alabi NO.pdf Download (835kB)

Abstract

In this paper, we have considered mainly four of the most common computationally efficient linear and non-linear classification kernels (i.e. support vector classifier or SVC, polynomial basis kernel, radial basis kernel and sigmoid kernel) called the support vector machines (SVMs) to model the direction of exchange rate using crude oil price and inflation rate in Nigeria between 2004 and 2017. This study aims at testing the SVMs accuracies and forecast strength on the direction of exchange rate in Nigeria data. We based our study on 167 observations collected from the database of the Central Bank of Nigeria (from 2004 to 2017) which were divided into two parts; training dataset (2004 to 2007) and test dataset (2008:2017). One-against-one approach and 10-fold cross-validation resampling technique were used to select amongst several cost parameters C and  values for the linear and non-linear basis kernels. Specifically, cross-validation selected C = 0.5, 105, 0.1, 103 for the SVC, polynomial basis kernel of degree = 2, sigmoid kernel and radial basis kernel respectively. For the radial basis kernel,  = 0.001 was optimal amongst 153 positive  values implemented in the cross-validation procedure. A total of 155, 148, 158 and 150 support vectors were generated for the SVC, polynomial basis kernel of degree = 2, sigmoid kernel and radial basis kernel respectively. All kernels except the radial basis kernel performed slightly better on the test dataset. However, the confusion matrices showed that only the polynomial basis kernel of degree = 2 classified one of the three observations in the “unchanged” class correctly on the test data set. Packages from various R libraries were deployed throughout the paper.

Item Type:	Conference or Workshop Item (Paper)
Subjects:	Q Science > Q Science (General)
Divisions:	Faculty of Engineering, Science and Mathematics > School of Chemistry
Depositing User:	Mr Taiwo Egbeyemi
Date Deposited:	09 Jun 2020 16:04
Last Modified:	09 Jun 2020 16:04
URI:	http://eprints.federalpolyilaro.edu.ng/id/eprint/383

Actions (login required)

View Item