Induction Of Classification Rules From Noisy Sonar Features

Abstract

Developing rules to classify surface ships and submarines based on noisy features extracted from underwater acoustic signals is difficult and time consuming; indeed, it is difficult to generate decision making rules for any complex system. The Operational Evaluation Modeling (OpEM) Simulation Tool Kit including Expert System Controller and Induction Program, discussed in this paper, greatly mitigates this problem for most complex system simulation projects such as airports, factories, or space defense systems. A theory for inductive learning of decision making rules in cybernetic systems is presented to provide a concise way-of thinking about inductive learning. An OpEM directed graph model is presented that describes operation of a sonar detection and classification system. An OpEM Pascal simulation of this system demonstrates how effective classification rules can be that are induced from simulation generated cases, even with noisy features present. The OpEM induction program is described and innovative features are discussed. Rules generated by the OpEM induction program are compared to rules obtained by Ross Quinlan's ID3 induction program. Performance of the OpEM induction program when noisy features are present is compared with ID3.

Simulation-Based Engineering Of Complex Adaptive Systems

“…The Operational Evaluation Modeling (OpEM) graphical discrete event simulation language [12,13] was developed to describe the context-sensitive transitions required for CAS interactions. The challenge in developing the graphical version of this language was to apply lessons learned from experience obtained from the OpEM Pascal Simulation programming system [2,[12][13][14][15][16][17][18][19] used during the past 15 years. This system, though flexible in terms of data structures and event action code that can be created, is not easy to program.…”

Optimization of Simulated System Effectiveness Using Evolutionary Algorithms

“…This state-of-the-art graphical, discrete event simulation library works with ExtendTM (by Imagine That, Inc.) and has the power of much more costly, general-purpose simulation products. OpEMCSS, which was developed based on earlier work [5][6][7][8][9][10][11][12][13], is capable of representing several commonly used systems views including queuing, process and resource, and message flow models (see http://ecs.fullerton.edu/~jclymer~.…”