Different scholars have recently advocated the use of so-called realized volatility measures constructed from the summation of high-frequency intra-daily squared returns as an accurate measure for true volatility. Scholars have also advocated that, as a model for realized volatility forecasting, the HAR model performed better. This paper analyzes the daily realized volatility calculated using 5-minute returns of the Nikkei 225 stock index. The purpose of this paper is to compare the forecasting performances of HAR model by using classical, robust and support vector regression techniques.