AN APPLICATION OF MULTIVARIATE GAUSSIAN DISCRIMINANT CLASSIFIERS TO THE DIRECTION OF INFLATION RATE IN NIGERIA

Alabi, Nurudeen Olawale and Are, Stephen Olusegun (2017) AN APPLICATION OF MULTIVARIATE GAUSSIAN DISCRIMINANT CLASSIFIERS TO THE DIRECTION OF INFLATION RATE IN NIGERIA. In: School of Information and Communication Technology (SICTCON), Auchi Polytechnic, 2017, The Federal Polytechnic, Ilaro.

Text Conference 4_Alabi NO.pdf Download (1MB)

Abstract

In this research work, we have considered two different statistical learning techniques for classification such as Linear Discriminant Analysis (LDA), and Quadratic Discriminant Analysis (QDA) in modeling and forecasting the direction of inflation rate in Nigeria between 2004 and 2016 using two macroeconomic variables such as the nominal exchange rate and crude oil price. The direction of inflation rate was classified into k = 3 levels i.e. “Down”, “Unchanged” and “Up”. Analysis revealed that during the period under review, inflation rate rose in 86 months, decline in 66 months and remained unchanged for 4 months out of the 156 months with prior probabilities of 0.4231, 0.0256 and 0.5513 respectively. LDA on the direction of inflation rate was based on the assumption that the two predictors are multivariate normally distributed with common covariance matrix. LDA classifier was based on two linear discriminant functions. The between- and within-class standard deviation ratios for the two linear discriminant variables generated are 2.2299 and 0.0826 respectively. The LDA output shows that whilst about 53.20 per cent of the training observations were correctly classified into “Down”, “Unchanged” and “Up” categories, 46.80 per cent were misclassified. Hence the accuracy of the LDA classifier is 53.20 per cent. In addition to the assumption of the vector of expanatory variables (i.e crude oil price, exchange rate and interest rate) coming from a multivariate Gaussian or normal distribution, QDA estimates a separate covariance matrix for each class. The QDA output shows that 40.38 per cent of the training observations were incorrectly classified while 59.62 per cent of the observations are correctly classified. Based on a specific economic scenario whereby a test dataset of observations between 2015 and 2016 was used, the performances of the classifiers revealed a test error rate of 95.83 per cent for the LDA and 8.33 per cent test error rate for QDA. R programming language packages were employed throughout the analysis.

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 16:05
Last Modified:	09 Jun 2020 16:05
URI:	http://eprints.federalpolyilaro.edu.ng/id/eprint/384

Actions (login required)

View Item