Dictionary · Glossary of Computer System Software Development Terminology

For a computer to process a set of data whose attributes have been divided into two groups and derive a relationship between the values of one and the values of the other. These two groups are sometimes called predictor and targets, respectively. In statistical terminology, they are called independent and dependent variables. Respectively. The learning Is "supervised because the distinction between the predictors and the target variables is chosen by the investigator or some other outside agency.

The accuracy of a machine learning system is measured as the percentage of correct predictions or classifications made by the model over a specific data set. It is typically estimated using a test or "hold out" sample, other than the one(s) used to construct the model. Its complement, the error rate, is the proportion of incorrect predictions on the same data.

A proposed method for modifying machine learning algorithms by allowing them to specify test regions to improve their accuracy. At any point, the algorithm can choose a new point x, observe the output and incorporate the new (x, y) pair into its training base. It has been applied to neural networks, prediction functions, and clustering functions.

Systems that classify or predict a concept (e.g., credit ratings or computer intrusion monitors) over time can suffer performance loss when the concept they are tracking changes. This is referred to as concept drift. This can either be a natural process that occurs without a reference to the system, or an active process, where others are reacting to the system (e.g., virus detection).

The process of identifying homogeneous subgroups within a data table.