I. Introduction
Electricity enriches our lives. The electric power company estimates the demand for the electric power every day. For example, in the case of TEPCO(Tokyo Electric Power Company) Power Grid, the date and time of daily peak load demand are disclosed in Electrical Forecast [1]. Since electric power cannot be stored, it is important to reduce surplus electric power for the sake of the environment problem as prevention from global warming etc.. In previous studies, the authors have confirmed that meteorological conditions are important for load demand forecasting and that determining forecast values with a single forecast model is a risk. In this paper, we have proposed a prediction method combining machine learning methods and statistical methods that is different prediction characteristics. Furthermore, we consider the effective input variables for daily peak load demand forecasting.