ON PATTERN CLASSIFICATION ALGORITHMS--INTRODUCTION AND SURVEY

THE SPECTRUM OF EFFORT IN PATTERN CLASSIFICATION IS BOUNDED ON ONE SIDE BY THE ENGINEERING ASPECT, WHICH IS MAINLY CONCERNED WITH THE IMPLEMENTATION AND DESIGN OF ACTUAL RECOGNITION DEVICES, AND ON THE OTHER SIDE BY THE ARTIFICIAL INTELLIGENCE ASPECT, IN WHICH PHILOSOPHICAL QUESTIONS OF LEARNING AND INTELLIGENCE ARE DEALT WITH. RATHER THAN TREATING EITHER OF THESE TWO EXTREMES, THIS PAPER SURVEYS THE ANALYTICAL ASPECTS OF THE PATTERN CLASSIFICATION PROBLEM. THE TECHNIQUES OF ESTIMATION, DECISION, AND OPTIMIZATION THEORY ARE BROUGHT TO BEAR ON THIS PROBLEM, WHICH IS VIEWED AS ONE OF MAKING DECISIONS UNDER UNCERTAINTY. PATTERN CLASSIFICATION CAN BE DIVIDED INTO THE TWO FUNDAMENTAL PROBLEMS OF CHARACTERIZATION AND OF ABSTRACTION AND GENERALIZATION. THIS SURVEY CONCENTRATES ON THE VARIOUS ALGORITHMS FOR THE SOLUTION OF THE ABSTRACTION PROBLEM, AND THE AUTHORS CONCLUDE THAT CLASSIFICATION ALGORITHMS CAN BE BROADLY DIVIDED INTO TWO GROUPS, PROBABILISTIC AND DETERMINISTIC. TWO OTHER CONCLUSIONS ARE THAT CHARACTERIZATION REMAINS A MAJOR OPEN PROBLEM AND THAT A RELATIVELY SMALL AMOUNT OF EXPERIMENTATION WITH THE ALGORITHMS HAS BEEN CARRIED OUT WITH REAL-LIFE CLASSIFICATION PROBLEMS. /AUTHOR/

• Supplemental Notes:
  • Vol 56, pp 2101-2114
• Authors:
  • Ho, YU-CHI
  • Agrawala, A K
• Publication Date: 1968-12

Media Info

• Serial:
  • Ieee Proceedings

Subject/Index Terms

• TRT Terms: Algorithms; Decision theory; Pattern recognition systems
• Uncontrolled Terms: Patterns
• Subject Areas: Highways; Operations and Traffic Management;

Filing Info

• Accession Number: 00225676
• Record Type: Publication
• Source Agency: Traffic Systems Reviews & Abstracts
• Files: TRIS
• Created Date: Jul 21 1970 12:00AM