Results show that FM has excellent performance. Error probability of both Least Square model and Crossvalidation estimation are all under2.5%. FM is a kind of good supplement to traditional classifiers.

• 结果表明；分类效果较好；无论是总体的模型还是交互检验；预测误差率都在2.；5%25下；fm变量是传统分类变量的有益补充。