APPLICATION OF LINEAR REGRESSION MODEL (LRM) FOR THE PREDICTION OF PARTICULATE MATTER (PM2.5) IN MARBLE PROCESSING COMMUNITY, IKPESHI AND ENVIRONS, EDO STATE.

ABSTRACT

This study investigated the use of Multiple linear regression (MLR) model to predict the concentration of PM2,5  in Ikpeshi and environs, Edo state. Air quality monitoring data including temperature, wind speed, humidity and PM2.5 were collected from nine (9) sampling points in the study area. Preliminary analysis such as homogeneity test, normality test, outlier detection test and reliability analysis were conducted before modelling to ensure the suitability of the acquired data. The Jarque-Bera, Shapiro-Wilk and Kolmogorov-Smirnov tests were employed in the normality test and they all indicated normal distribution for humidity while temperature wind speed and PM2.5 were not normally distributed. No outliers were detected using labeling rules and boxplots. Residual mass curves confirmed homogeneity of the datasets and Cronbachs alpha (0.83) along with ANOVA indicated high reliability of the data. Multiple linear regression (MLR) was employed using temperature, windspeed and humidity as independent variables to model PM2.5 concentrations. The diagnostic tests for MLR reveal the presence of heteroskedasticity in the data since the computed (P value) based on F statistics and langrange multiplier is greater than 0.05. The presence of heteroskedasticity in the data implied that linear regression may not be the best model to assess the relationship between independent variables and the dependent variable PM2.5. The test also indicated that there was no serial correlation in the data as the Durbin- watson value was less than 2, this validates the suitability of multiple linear regression for establishing the relationship between the dependent and selected independent variables . The variance of inflation factor was also calclated and based on this calculation it was found that the selected independent variables have a centered VIF less than 10 which indicates an absence of multicollinearity. To check the efficiency of the developed model an output of regression analysis was conducted, which revealed An F value less than 0.05 which means the model can effectively determine relationship between the dependent high standard error and independent variables and a high standard error which limits its application. As a result it could be inferred that Multiple linear regression could determine the relationship between the independent variables (wind speed, temperature and the dependent variable (PM2.5)