C4.5

What it is

C4.5 is a machine learning system for building classifiers from data. It starts with large sets of cases that belong to known classes and are described by nominal and numeric properties. From those cases it produces patterns expressed as decision trees or as sets of if-then rules. The stated aim is not only accuracy but also models that people can understand.^[1]

A decision tree is a flowchart-like model: it asks a series of questions about properties and routes each case toward a class. A set of if-then rules expresses similar knowledge as conditions and conclusions. Both forms are meant to be readable, so a person can inspect why a case was assigned to a class. That readable output is part of what the book emphasizes.^[1]

Origin and documentation

The system is documented in J. Ross Quinlan's book *C4.5: Programs for Machine Learning*. ACM's Digital Library records the book as published on 1 January 1993 by Morgan Kaufmann. It is a 302-page complete guide to C4.5 as implemented in C for the UNIX environment. The book includes a guide to using the system, about 8,800 lines of source code, and implementation notes.^[1] ScienceDirect lists the book as published in 1993 under the Morgan Kaufmann imprint, with ISBN 978-0-08-050058-4.^[2]

What the book covers

The chapter list gives a view of the system's scope. It includes Constructing Decision Trees, Unknown Attribute Values, Pruning Decision Trees, From Trees to Rules, and Windowing.^[2] These topics correspond to the basic steps of turning a table of cases into a usable classifier: growing a tree, dealing with missing or unknown property values, simplifying the tree by pruning, and converting the tree into rules. The book also treats windowing as a separate topic. The emphasis on understandable models matters because a decision tree or a set of if-then rules can be read by a person, not only executed by a program.^[1]

Unknown attribute values matter because a case may not record every property. Pruning matters because a tree grown too finely may fit quirks of the cases rather than general patterns. Converting trees to rules gives users another way to inspect and deploy the learned knowledge. The chapter list shows that these practical steps are treated alongside tree construction.^[2]

Use and position

ACM says the system had been applied successfully to tasks involving tens of thousands of cases described by hundreds of properties.^[1] That scale is part of why the book is more than a short description: it includes the source code and implementation notes, so readers can study how the method was made to work in practice. C4.5's place in the development of machine learning is tied to this combination of a working implementation, a published account of its use, and output forms that are meant to be understood. The 1993 book made the system available as both a tool and a case study in implementation.^[1][2]