(a) (b)
he Gamma densities estimated for two types of tumours using the parametric
or the breast tumour area feature of the breast cancer data. (a) The malignant
sity. (b) The benign tumour density.
e non-parametric approach
gram approach is generally used for generating a visible density,
loyal to the true data distribution. But, it cannot be used for a data
high-dimensional space. The parametric approach enjoys the
y and a resulting density model is parsimonious in structure. It is
d on two important assumptions. First, the density format is known
e. Second, the density is a single-mode data distribution such as a
ussian or a single Gamma distribution. Therefore to estimate the
or a more complicated distribution, the alternatives should be
d. Among them, the non-parametric approach is the one.
dvantage of the non-parametric density estimation approaches is
can estimate a density function without a priori knowledge about
stribution looks like and still maintain a limited flexibility of the
risation of a density model [Rosenblatt, 1956; Parzen, 1962;
al., 2000].
he kernel method
e several non-parametric density estimation approaches. The
ethod is one of the widely used ones. It employs a set of raw data