Page 263 - Contributed Paper Session (CPS) - Volume 7

CPS2084 Siti Aisyah Mohd Padzil et al.

MA approach differs from MS as there is no elimination of insignificant
variables in the model-building. Best model of MA will include all covariates
being studied in final model even if it’s insignificant. As an alternative
approach, AMA will produce final model which comprises of only significant
variables as minimization of covariates in the final best model will results in
numerical stability and reliable results (Bursac, 2008).
Based on Figure 1, the process of AMA is similar with MA except for
elimination of insignificant variables. After model was obtain from the
combination of computed coefficients, independent variable with -value
larger than 0.05 which indicate insignificant variable will be eliminated one by
one. This process only allows one insignificant variable to be eliminated at a
time and after a covariate is remove, the process of obtaining model will be
rerun until best model with only significant variable is obtain.
To compare the accuracy of predictive model produced by traditional MA
and alternative method, Root Mean Square Error (RMSE), Mean Absolute Error
(MAE) and Mean Square Error (MSE) are computed. Equation 12, 13 and 14
presents the formula for prediction error as suggested by (Chai and Drexler,
2014).

250 | I S I W S C 2 0 1 9