In this paper, a new optimization methodology is proposed that can be used when an accumulated data is given. The proposed methodology consists of two methods; the performance prediction method and the optimization method. The existing methodologies using an accumulated data have their limits in predicting the performance accurately at a point of interest. To enhance the performance prediction accuracy, local metamodeling approach in which we use only the data points surrounding a point of interest instead of the entire accumulated data is proposed in this study. As for the optimization method, the sequential approximate optimization concept is adopted to utilize the local metamodeling approach effectively. The efficiency of the proposed methodology is demonstrated by comparing its effectiveness and accuracy to those of existing methodologies using various accumulated data.