Glossary of terms

The percentage of correct predictions a model makes for categorical decisions. This is more complex for numerical decisions.

A machine learning (ML) technique which asks a human-in-the-loop to provide outcomes for specific inputs. These inputs are used by the machine to optimize the rate of learning and minimize the number of interactions with the human-in-the-loop. See also Guided learning.

A machine learning system that requires no user input to fit data, analyze the relationships between the variables, and make predictions on new data. A typical example is the Automatic Statistician.

A categorical variable that takes one of two values, typically true or false.

The process of reviewing model inputs and outcome predictions, and overwriting model decisions if required.

A variable that takes one of a fixed set of values. If the values are unordered then it is a “nominal” categorical variable, otherwise if the values are ordered it is an “ordinal” categorical variable.

See Confidence.

Part of a rule. All the rule's conditions must be satisfied in order for the rule to apply. For example: stock room = full. Also known as 'constraint'.

This defines how sure a model is in its prediction, for example how many times out of 100 it expects to be correct. Also known as 'certainty'.

The type of a variable in a problem. Can be categorical or numerical.

See Outcome variable.

See Rule.

A machine learning technique which asks a human-in-the-loop to provide inputs that generate specific outputs. These outputs are used by the machine to optimize the rate of learning and minimize the number of interactions with the human-in-the-loop. See also Active learning.

A human that provides guidance to a machine learning method or automation.

Input to the decision making process.

A relative value that a model gives to an input variable compared to the other input variables to determine its importance. A higher value indicates that the model treats that variable as more important. Some models permit the user to alter these weights. Also known as feature importance or feature weight.

A machine learning or statistical non-parametric model that makes predictions (categorical or numerical) for an input based on the k nearest neighbors (kNN) to the input in the training data. The nearest neighbors can be considered equally (an unweighted kNN model) or weighted by their inverse distance or similarity (a weighted kNN model).

Systems in which machines learn by themselves.

A machine learning system that predicts an outcome (decision) based on a given a set of features.

A subset of the samples that are considered to be similar to the values that have been entered for the variables. See kNN model.

Systems that can be configured without programming. These are typically designed for a business user rather than an engineer to use.

An unordered categorical variable. For example, “cat”, “horse”, “dog”.

A machine learning model that makes few assumptions of the structure of the problem and fits to the data. Typically, these models are more flexible than standard statistical parametric models.

A variable that takes any value in a range. For example, 1, 1.2, 100, -3.

An ordered categorical variable. For example, “small”, “medium”, “large”.

The outcome of the decision making process. Also known as 'decision'.

A user-defined exception to the Decision model. This is a list of conditions or constraints that when satisfied force the outcome to a particular value. Also known as 'exception'.

A data point that is one complete set of input variable values possibly also associated with an outcome variable value.

See kNN model.

Generic term for either an input or outcome (decision) variable.