Title

The Use Of Decision Trees As A Reliable Management Tool

Abstract

The complexity of decision making involving total system reliability is significantly affected by the number of subsystems present in the system configuration. Also, the number of operational subsystems influences the dependency relationship which greatly impacts total system performance. Often during traditional decision analysis paradigms, to reduce the complexity of the analytic analysis each subsystem is considered to be independent. This research exploits the use of decision tree analysis to determine dependency structures that are useful in analyzing overall system reliability as an aggregate construct of subsystem analysis. In today's design environment of complex system structure and architectures, managers and engineers want to feel more confident regarding total system design decisions. Decision-making in this environment is not an easy task since systems are becoming more complex as the number of subsystems to be integrated continues to increase. It is estimated that most of the systems in the future will be comprised of complex systems that violate the traditional theoretical construct, which assumes the subsystems or components are independent and equally weighted. Thus, for future analysis the appropriate construct paradigm must address facts that clearly distinguish that the components of complex systems are dependent, and these subsystems are usually not equally critical or weighted when determining total system performance. Therefore, the goal of this research effort is to explore the formation of valid, accurate, and reliable paradigms for decision-making through using decision trees while considering the interrelationships of decision analysis and reliability constructs. The application of this research approach ensures confidence, support, assuredness, and the control of decisions that negatively influence system performance and compromise system reliability. The specific efforts explored to meet the stated goal of this research consist of identifying, selecting, analyzing the component, subcomponent, and factor variables that most strongly define the relationship with subsystems dependency and reliability. Also, research efforts include the testing and evaluation of the decision tree to quantify the relationship between subsystems dependency and reliability. There are two expert approaches for using decision trees: Quinlan's approach (Quinlan, 1993), and Breiman's approach (Breiman et al., 1984). Based on these approaches, there are three steps involved in the development of decision trees: the tree-building step, the tree-pruning step, and the tree-validation step. Since a decision tree is a classification exercise with a given data (input/output) a function can be created to establish a relationship between inputs and outputs. The tree-building step consists of designing a tree with the given data and always starting from the top, and the purpose is to create smaller data samples of single nomination. Pruning permits the prioritization and simplification of the tree from the building stage. The purpose of the pruning step is to simplify and facilitate the work on the tree. The validation stage is based on selecting the best-performed pruned tree of the original tree. Decision trees have been selected as one of the best decision sciences methods since its origins. Once the decision tree is constructed, the system is expressed in terms of reliability as a complex system. This will prove that the interdependency of subsystems greatly influences the decision over the whole system. This research presentation discusses the decision tree development process, explores examples of its use, and the decision-making alternatives resulting from this approach.

Publication Date

12-1-2004

Publication Title

IIE Annual Conference and Exhibition 2004

Number of Pages

1957-

Document Type

Article; Proceedings Paper

Personal Identifier

scopus

Socpus ID

30044447812 (Scopus)

Source API URL

https://api.elsevier.com/content/abstract/scopus_id/30044447812

This document is currently not available here.

Share

COinS