Efficient and effective data classification is one of the most fundamental problems of computer science due to a huge number of important applications. As such, there have been many models that are capable of doing some form of data classification using a large number of varied algorithms and architectures. However, few share properties that implicitly take advantage of the distributed environment where classification has its most potential.

Therefore, to resolve this dilemma in this thesis a generic classification architecture com posed of mutually adaptive distributed heterogeneous agents was designed, implemented and experimentally analyzed. The system defined creates a multi-agent distributed ar chitecture whereby individual agents cooperate and collaborate with each other to autonomously perform classification on arbitrary data using a hierarchical paradigm.

The main contributions of this work are the following: agents follow hierarchical pro cesses to form an organization which can autonomously train, test, assign and evaluate work of other agents. Through experimental analysis it was found that this design improves the classification accuracy relative to previously implemented classification algorithms and architectures. Furthermore this analysis shows that the generated architecture will au tonomously adapt to previously unlearned data and responds to failures with little or no degradation of classification quality.

Library of Congress Subject Headings

Intelligent agents (Computer software); Discriminant analysis; Automatic classification; Computer algorithms; Data mining

Publication Date


Document Type


Student Type


Degree Name

Computer Science (MS)

Department, Program, or Center

Computer Science (GCCIS)


Leon Reznik

Advisor/Committee Member

Roger Gaborski

Advisor/Committee Member

James Heliotis


Physical copy available from RIT's Wallace Library at QA76.79.I58 H37 2007


RIT – Main Campus