3.2 ML categories
ML methods are broadly divided into supervised, unsupervised, and
semi-supervised learning [93]. Unsupervised learning aims to uncover
the hidden patterns and deduce the structures of unlabeled training data
[18]. Unsupervised learning approaches cluster subgroups of data
with similar properties or features into separate categories. Moreover,
dimensionality reduction methods mentioned earlier as preprocessing
operations are unsupervised learning methods that reduce the number of
old features and create new principal features with minimum information
loss [94]. However, because there are no target values in
unsupervised learning methods, they cannot build an independent
predictive model [95]. On the other hand, supervised learning is
applied to learn and discover associations relationships between
features and target values in a labeled dataset [96]. Therefore, the
built model can be used as a predictive one to test previously unseen
data. Two main supervised methods are classification for discrete class
labels and regression for numerical quantities [97]. It is
noteworthy that the output of unsupervised learning methods (e.g.,
low-dimension space features) can be used as the input of supervised
learning algorithms. For semi-supervised learning, there are both
labeled and unlabeled data in the training set; But usually, the amount
of unlabeled data is more. In this method, one can use labeled data to
create labels for unlabeled ones [98].