부산대학교 도서관

This paper presents an original method for estimating a density. The estimator is mixture of kernel estimators of components of an ``approximate normal mixture model'' for the data. The formula of the Asymptotic Mean Integrated Square Error is established. A simulation is conducted to assess the performance of the method. \par Although the simulation carried out in the paper suggests that the new method might be superior under certain circumstances, the method needs further studies to be considered as a sound alternative to classical approaches. The consistency properties are proven when the parameters of the mixing model are nonrandom. However, in practice (as in the simulation), they are to be estimated. Also, the selection of the number of mixture components is yet to be addressed.