This paper presents a new technique for model order reduction with dominant eigenvalue preservation. The new method is formulated based on the prediction of an artificial neural network along with the linear matrix inequality optimization technique. Utilising the non-projection state residualization technique, a reduced model order is obtained. Results' comparison with well-known techniques such as Schur decomposition, proper orthogonal decomposition, and state elimination through balanced realization shows the superiority of the new method.