Owing to the introduction of an energy-monitoring device called the "smart meter," a large amount of power-demand information regarding private houses is being stored around the world through the development of smart grids. Such electric consumption data include a variety of features related to the structure of the customer's home, as well as their family structure and lifestyle. To capture the meaningful features from the power-demand information, a proper extraction method should be considered such as a discrete Fourier transform (DFT) or discrete cosine transform (DCT). This paper proposes the use of two frequency analysis methods, i.e., DFT and DCT, toward power-demand information. The converted power-demand information is transformed using three expression methods. A power-demand dataset was gathered from more than 100 houses in Yokohama, Japan, over a two-year period. The numbers of feature values in these three expression methods were compared by estimating the floor space of each customer's house with two support vector classification (SVC) values using both linear and RBF kernels. As a result, the proposed expression method performed better than the other two feature expression methods. The mean accuracy of this expression method is better than the other two by 20%, and the highest level of accuracy in terms of the floor-space estimation is greater than 95% when using liner SVC (L2-SVC). The accuracy shows a distribution of 50% to 100% regardless of the number of selected frequencies transformed using a DFT and DCT.