GSJ: Volume 8, Issue 10, October 2020, Online: ISSN 2320-9186 www.globalscientificjournal.com Improving the Performance of Linear Regression Model: A Residual Analysis approach Joshua Hassan Jemna 1* , Kazeen Eyitayo Lasisi 1 Emmanuel Alphonsus Akpan 1 , Alhaji Gwani Abdullahi 2 Abubakar Abdullahi 3 , Akpensuen Shiaondo Henry 1 1 Department of Mathematical Sciences, Abubakar Tafawa Balewa University Bauchi 2 Department of Mathematical Sciences, Bauchi State University Gadau. 3 Mathematical unit, department of General Studies, Gombe State Polytechnic Bajoga, Henroo85@gmail.com ABSTRACT: The study considered GDP as dependent variable, agriculture, industry and trade as independent variables respectively. The data was gotten from the central bank of Nigeria from 1983 to 2019. The aim of the study was to apply residual analysis approach to improve the performance of linear regression. The relationship between the dependent variable, GDP, and the independent variables, Agriculture, industry and trade was determined using the ordinary least squares estimation method. The results of the ordinary least squares estimated regression showed that Agriculture, industry and trade contributed significantly to GDP and were able to explain about 89% of the variance in GDP. Furthermore, evidence from Breusch- pagan test confirmed that heteroscedasticity exist in the residuals of the linear regression model while ACF and PACF revealed that the error terms were autocorrelated. The Jarque- Bera normality test revealed that the errors were normally distributed. To account for the autocorrelation in the error terms, we applied two different generalized least squares models, that is regression ARMA model (RAM) and overfitted regression ARMA model (ORAM) with different ARMA components, that is, ARMA (1, 2) and ARMA (1, 3) respectively. Results of our analysis revealed that the estimates of the RAM model were better than those of ORAM. Also based on minimum information selection criteria (AIC, BIC, LOGLIK) RAM was selected as the suitable model. The autocorrelation in the error terms was found to be completely modelled by ARMA (1, 2) process. An ARMA (1,3) model (specification) would be unusually complicated, but in any event the tests support the ARIMA(1,2) specification. Keywords: GDP, Agriculture, Industry, Trade, ARMA Generalised least squares. GSJ: Volume 8, Issue 10, October 2020 ISSN 2320-9186 212 GSJ© 2020 www.globalscientificjournal.com