Machine Learning is a computer science field which aims to study statistical algorithms and models that allows software to perform tasks without being explicitly programmed. Contrary to classical software where all instructions must be provided, machine learning algorithms construct (train) statistical models based on sample data (so-called training data). The statistical model is used to compute outcomes for new data samples.

Machine Learning is a branch of artificial intelligence. Subfields of machine learning include:

• Data Mining, which uses unsupervised methods for better data understanding
• Deep Learning, which constructs Neural Networks models with many (thus, deep in the name) layers

Supervised learning

The process of building a statistical model based on data which contains an input feature vector (vector of numbers) and the desired output value (target column). The data used for model construction is known as training data. Each data sample from training data should have value for the target column. Trained model can be used to infer (predict) outputs for new data samples which were not used for training - the new data is so-called testing data.

The examples of supervised learning are:

• classification - task where model’s output is assigned to limited set of values. For example, in case of binary-calssification the output values have two values (usually, 0 and 1).
• regression - task where target column can be any number within a range.
• similarity learning - task of finding related samples, with application to recommendation systems

The some examples of supervised algorithms:

• Decision Tree
• Linear Regression
• Neural Networks
• Random Forest

Semi-supervised learning

In the case of supervised learning all samples have values for target column. However, in some situations obtaining all target values is impossible. In such cases, semi-supervised methods can be applied - it is technique of model training with missing target values.

Unsupervised learning

The unsupervised learning describes types of algorithms that are searching for structure in the data without any guidance (supervisory target column). It is used in cluster analysis to define related subsets of samples in the data - clusters.

Reinforcement learning

Reinforcement learning is a task of learning agents to take such actions which maximize the cumulative reward. The learning process is made by attempting many times to complete the process by the agent. The steps with undesired outcomes are penalized whereas actions with positive outcomes are rewarded.

Who is using machine learning?

Examples of machine learning use cases:

• Financial services - businesses in the financial industry uses machine learning for fraud detection, credit scoring, investing opportunities identification, automatic trading, customers targeting
• Retail uses ML for items recommendation, churn prediction, personalized market campaigns, customer insights
• Health care uses data from past treatments to predict harmful events before they happen
• Energetic sector - Machine Learning is used for electricity trading, new energy source discovery
• Insurance - Machine Learning models are used for policy scoring, risk assessment, and management