شبكة المعلومات العربية التربوية

The current research objective is to study the presence of outliers on multiple-linear regression model and its impact on prediction’s accuracy and it show they are treated via the model’s parameters. The qualitative approach was used in this study. A sample of 60 students selected from population of 204 students of faculty of engineering for academic year 1436-1437 ah. Multi-linear regression model and some methods and tests of diagnosis and processing the outliers were applied. Accordingly, outlier values were detected visually using box and whisker plot or box plot. Main diagonal elements of the hat matrix were also used to detect outlier values in the independent variables, and in addition to that using studentized deleted residuals to detect them in the dependent variable. For the effect of outliers, it was detected using the following methods: difference in fits (DFFITS) measure, difference of beta values (DFBEATS) measure (effect on regression coefficients), cook's distance measure for measuring the effect on all regression coefficients, and covariance ration (COVRATIO) to measure (effect on standard errors).outlier values were processed using the methods of deleting and amputation average, with measuring the model leverage pre-and post- processing via using some indicators such as: R square (R 2 ), R square adjusted and f test, mean absolute deviation (MAE), root mean square error (MRSE), and mean absolute percentage error (MAPE), in addition to Theil's inequality coefficient. The research concluded with various findings, the most important ones are: there is a significant impact of outlier values on parameters of the multi-linear regression model, where they amplify the mean squared error (MSE), and reduce r 2 value and f value. The research has concluded also that the outlier values processing by amputation average is better than deleting them. The study recommended that: it is necessary to diagnosis and reduce the Outliers effect when applying the regression analysis model to obtain better model. (Published abstract)