Categories of Machine Learning Techniques

   Machine Learning refers to create and apply models by learning data. Inside rules, correlations, and the rational logics in a data are discovered by machine learning algorithms to predict or make the decision for new inputs. Data is the primary resource for machine learning algorithms. Understanding data structure helps to choose the right learning algorithm.
   In this post, I briefly explain machine learning methods based on data content.

   Supervised learning. A dataset contains an observation data with inputs and their known outputs.  Based on those input and output values, the algorithm builds the best possible function that represents y through the x by iteratively learning a dataset. Supervised learning is used in classification and regression problems.

   Unsupervised learning. It is a set of data without output values/labels. Since there are no answers in a dataset, the algorithm cannot be supervised. Thus, it is called unsupervised learning. The algorithm learns the overall structure of data, performs clustering, anomaly detection, and association check task.

   Semi-supervised learning. A dataset contains input data with partially output values/labels. Algorithm labels unlabeled inputs by learning the labeled part of it. This approach is mixed with supervised and unsupervised techniques.

   Reinforcement Learning. The agent learns about the environment and actions through the interaction to increase its reward or score. The actions in a given environment become a training source for the model. This approach is used in the areas of games, robotics, optimal control, and others.