different clusters and generate meaningful interpretations.
most popularly used estimation approach for the mixture model
m is the expectation-maximisation (EM) algorithm [Dempster,
77], which is a two-step iterative procedure (the E step and the M
rting from parameter initialisation using random values. The
ties are estimated in the E step using the information of current
rameters, i.e., the cluster centres, the cluster covariance matrices
mixing coefficients. If model parameters have not yet been
d, information of probabilities are used to update model
rs in the M step. In other words, new model parameters are
d in the M step based on the current cluster membership function
hich are the probabilities.
mixture model algorithm has been widely employed for analysing
l/medical patterns. For instance, it has been used to improve the
of the infection prediction when investigating the Bovine viral
infection among cattle herds [Eze, et al., 2019]. The study
some interpretable epidemiologically clusters. Another recent
s employed the Gaussian mixture model for breast cancer
based on mRNA expression data which contains over 300
amples of breast cancer and healthy tissue [Prabakaran, et al.,
he study has demonstrated a significant value of a Gaussian
model for clinical practice. The Gamma mixture model has also
ployed in an image analysis for developing unbiased pre-
g system of biology images [Llera, et al., 2019] and in a HIV test
ysis project [Rice, et al., 2019].
e 2.35(a) shows a data which is a mixture of three Gaussian
Table 2.10 shows the cluster result of two algorithms on this data
otal accuracies were 88.5% and 90.5% for models constructed by
eans algorithm and the mixture model algorithm, respectively.
35(b) shows a data set which is a mixture of three Gamma
The clustering performance is shown in Table 2.11. The
s were 96.9% and 97.2% for the models constructed by the K-
gorithm and the mixture model algorithm, respectively.