Purchasing actions on e-commerce sites have become very common for general consumers in recent years. Products that were used to be bought at offline shops are purchased are also handled. Such products, like gifts or durable consumer goods, are often purchased infrequently and whose prefer items change each time they are purchased. A lot of methods are proposed for analysis purchase history data in order to improve customer satisfaction. However, most of them focus on the co-occurrence relationship between customers and products and treat products purchased by the same customer as similar. Then, it is difficult to use the conventional product analysis methods that have been proposed for purchase history data is difficult for some kinds of data mentioned before. Therefore, the authors have proposed an analysis method with extracting features of products by using the coefficients of binary classifiers that discriminates product purchases or not. In this study, we conduct experiments with artificial data in order to evaluate our method. Specifically, we verify how accurately the coefficients can be estimated and under what circumstances they can be estimated more accurately.