Classification of Machine Learning Algorithms

Supervised Learning: A machine learning algorithm is classified as supervised learning when the process of building a model to predict input and output relationships relies on known (input, output) pairs within the training set (Figure 1).

In supervised learning, a label refers to the output or target variable associated with each input example in a dataset. It represents the ground truth or correct answer that the machine learning algorithm aims to predict or classify. Labels are typically provided by human annotators or derived from existing data. For example, in a classification task to predict whether an image contains a cat or a dog, the labels would be “cat” and “dog”. In a regression task to predict housing prices, the labels would be the actual prices of the houses. The presence of labeled data is crucial for training supervised learning models to make accurate predictions on new, unseen data.

Unsupervised Learning: In certain algorithms, the training data comprises solely input data without corresponding outputs. While machine learning algorithms may not be capable of predicting outputs, they can still extract significant information from the data. Such algorithms are referred to as unsupervised learning.

Clustering algorithms serve as prime examples of unsupervised learning.

Semi-Supervised Learning: The demarcation between supervised learning and unsupervised learning can sometimes be ambiguous. There exist algorithms where the training set comprises both data points with complete inputs and outputs, as well as data points with only inputs. These algorithms are termed semi-supervised learning.