Barin N. Nag

Performance evaluation of neural network decision models

Recently, promising results with neural networks have been reported for two-group classification problems such as bankruptcy prediction and thrift failures. Such applications are usually characterized by unequal frequencies of the two states of interest. This creates a major obstacle to effective performance evaluation of various decision models. Critical issues affecting the comparison include training sample design and the use of an appropriate performance metric. This paper addresses these two issues by comparing the performance of neural networks with that of statistical models for the decision problem of identifying successful new ventures.

A decision-support model for investment decisions in new ventures

Abstract The decision to invest in new ventures is characterized by incomplete information, where some measures of firm performance are quantitative, while some others are substituted by qualitative indicators. Qualitative attributes are defined in a relative sense. We develop a decision support model for identifying successful new ventures. The model integrates quantitative and qualitative variables through the use of the Analytic Hierarchy Process (AHP). The decision model gains in predictive accuracy through the use of qualitative attributes, and AHP imparts robustness to the qualitative measures.

Object-oriented model construction in production scheduling decisions

Abstract The importance of rapid and automated model development for decision support is recognized in production scheduling applications, where problem instances are often similar to some general model but not necessarily consistent with it, and yet there is little of either time or modeling expertise available. In the modeling literature, there are few, if any, constructs of model development from component parts. Model construction is closely associated with the structure and representation of model information and with the abstraction of problem information from the user. Proceeding from a taxonomy of general production scheduling models, we develop a schema to represent model information in an object-oriented framework that relies on the definitions of natural entities, rather than on a collection of models from past experience. We show the interactions of user information with the model objects in the construction of a model to support a decision in a problem instance.

A neural network model to predict long-run operating performance of new ventures

The prediction of long-run operating performance of new ventures, known as Initial Public Offerings (IPOs), represents a challenging decision problem. Factors adding to the complexity of the problem include asymmetrically informed agents, incentive problems, and inability to specify functional relationships between variables. Research literature identifying determinants of long-run performance of new issues is limited. This study uses a data driven, nonparametric, neural network based approach to predict the long-run operating performance of new ventures. The classification accuracy of the neural network model is compared with that of a logit model. Methodological issues such as sample design and estimation of optimal cutoff probabilities for classification are addressed. The results suggest that the neural networks generally outperform logit models.

Decision Making, Dashboard Displays, and Human Performance in Service Systems

Service systems are often characterized by large components of human work and the need to make decisions based on human performance. Human cognitive limitations and the abilities of computers to compensate led to decision support systems DSS. While a computerized DSS fits the needs of human cognitive limits, the strengths of human cognitive abilities are often overlooked. Human performance is often monitored by task completion in terms of timeliness and accuracy. A failure of this is that cognitive feedback is generally not given to the operator until after the task. Dashboard displays that are already widely used in manufacturing and other operational applications give current performance information and can take advantage of human cognitive capabilities. This paper presents the concept of decision support in human performance by exploring the extension of the dashboard display concept to human performance monitoring as a cognitive feedback mechanism. Examples specific to the service sector are provided in the context of a Help Desk environment.

Modeling Agent Auctions in a Supply Chain Environment

Agent-based auction technology has revolutionized auction trading in the Supply Chain environment by reducing the cost of transactions, and by increasing the satisfaction factor in matching requirements of seller and buyer agents. In this article, we have considered methods of matching quantities of buyer and seller agents by cooperation, with a priority on the buyer’s requirements. The article discusses the architecture of the agent and the agent community when there is cooperative matching of volume. We present a Dynamic Programming algorithm to describe the agent’s decision process, and heuristic algorithms as the practical solution methodology. The results of a simple experiment show the improvement achieved by cooperation.

An Experimental Investigation into the Effectiveness of OOA for Specifying Requirements

The application of object oriented concepts (OO) to the requirements phase of information systems (IS) and software development has been adopted by many proponents of IS and software development methodologies. Although many claims have been made about the effectiveness of OO techniques for improving requirements analysis, very few experimental studies have been done to substantiate these claims. This paper addresses this gap in the literature by conducting an experimental study that attempts to validate the effectiveness of object-oriented analysis (OOA) by comparing it to structured analysis (SA) for producing requirements. We argue that the quality of the requirements specification can be measured and that measurement can be used to compare the effectiveness of OOA and SA. We present an overview of the basic models and principles associated with OOA and SA, a discussion of quality in requirements definition, and a detailed discussion of the research methodology used. A review of relevant research is also presented and directions for further research are suggested. Our findings suggest that the OOA methodology does not necessarily produce better requirements statements.

Information enhancement in data mining: a study in data reduction

Data mining can be a powerful tool for information extraction from large amounts of data. One of the techniques used to enhance the information extraction process is data reduction. Based on manufacturing industry data collected from US Economic Census, we use as an example the construction of a typology of inventory strategy according to Porters five forces model. This study shows that data reduction e.g., more aggregate data and fewer variables enhances the information extracted e.g., clearer patterns.

ESOM: An expert system for multicriteria scheduling

Abstract There exist situations in operations scheduling where the parameters of the scheduling decision are nonquantifiable, or appear as nonnumerical entries in a database. Scheduling techniques based on numerical algorithms may be difficult to use in such situations. In many service industries, the nonnumerical character arises from constraints or preference parameters associated with service tasks, service agents, and the business operation itself. The agent to task assignment problem is one of satisfying multiple objectives, by taking into account the various constraints and preferences inherent in such problems. In this paper, we describe the development of an expert system prototype called ESOM, that aims to automate the assignment decision process of a childcare referral agency. We discuss some key research issues pertaining to service scheduling in general, and to the childcare assignment problem in particular. ESOM provides useful guidelines to future researchers and practitioners interested in building expert systems for service scheduling. We demonstrate assignments made by our prototype through an application scenario.