Unsupervised Machine Learning
In unsupervised learning, the aim is to learn groupings in data
or reduce their dimensionality. Contrary to its supervised counterpart,
there are no known labels to predict. Unsupervised learning is often
used for clustering analysis. Here, the algorithm aims to describe the
data in a limited number of clusters or groups, where goodness-of-fit
tests determine the most parsimonious model. An
example30 is the discovery of asthma phenotypes based
on longitudinal wheezing patterns or clinical variables. Techniques for
unsupervised learning are latent class analysis (LCA), k-means
clustering, principal component analysis (PCA), and Multidimensional
Scaling (MDS). Recently, also semi-supervised learning has grown
in popularity, which aims to overcome the lack of sufficiently large,
labeled datasets and the tedious task of manual labeling. It leverages a
dataset of yet unlabeled data to improve the performance of a model that
is initially trained on labeled data.