Soft Computing Applications in Business

The literature suggests that an ensemble of classifiers outperforms a single classifier across a range of classification problems. This chapter provides a brief background on issues related ensemble construction and data set imbalance. It describes the application of ensembles of neural network classifiers and rule based classifiers to the prediction of potential defaults for a set of personal loan accounts drawn from a medium sized Australian financial institution. The imbalanced nature of the data sets necessitated the implementation of strategies to avoid under learning of the minority class and two such approaches (minority over-sampling and majority under-sampling) were adopted here. The ensembles outperformed the single classifiers, irrespective of the strategy that was used. The results suggest that an ensemble approach has the potential to provide a high rate of classification accuracy for problem domains of this type.

In many business applications, it is important to analyze correlations between currency exchange rates. In this Chapter, we develop a technique for bicluster analysis of the rates. Our method is able to extract coherent cluster patterns from a subset of time points and a subset of currencies. Experimental results show that our method is very effective. The bicluster patterns are consistent with the underlying economic reasons.

Artificial societies are computer models in which the collective behavior of a population of simulated human decision makers is observed over time. Here we describe an artificial society of “soft computing agents” (model consumers) making probabilistic purchasing decisions about new technological products that are introduced by competing firms. The model is studied under varying conditions to determine the relative success of these firms as they pursue different pricing strategies designed to increase their market share. We find that a critical factor in determining the success of different pricing strategies is the utility that an individual consumer gains from other consumers adopting the same technology. Further, financial success is uncoupled from market share under some conditions, so (for example) while an inferior technology may gain substantial market share by aggressive price-cutting, it is unlikely to gain financial rewards. These results add to growing evidence that artificial society models may prove useful in improving our understanding of collective decision making in complex sociological, economic and business management situations.

Business risk identification is one of the most important components in business risk management. In this study, a knowledge ensemble methodology is proposed to design an intelligent business risk identification system, which is composed of two procedures. First of all, some data mining and knowledge discovery algorithms are used to explore the implied knowledge about business risk hidden in the business data. Then the implied knowledge generated from different mining algorithms is aggregated into an ensemble output using an evolutionary programming (EP) technique. For verification, the knowledge ensemble methodology is applied to a real-world business risk dataset. The experimental results reveal that the proposed intelligent knowledge ensemble methodology provides a promising solution to business risk identification.

This chapter investigates the appropriateness of the application of fuzzy decision trees on the evaluation of company audit fees, with attention to the sensitivity of the results. With the rudiments of fuzzy decision trees in a fuzzy environment, it implies a linguistic emphasis on the concomitant analysis, allowing readability in the fuzzy decision rules constructed. Two processes for the construction of membership functions (MFs) used to define the linguistic terms characterising the linguistic variables considered allowing the impact of considering alternative MFs. The tutorial fuzzy decision tree analysis clearly allows the construction processes to be exposited.

This chapter centres on a novel classification technique called NCaRBS (N-state Classification and Ranking Belief Simplex), and the analysis of Moody’s Bank Financial Strength Rating (BFSR). The rudiments of NCaRBS are based around uncertain reasoning, through Dempster-Shafer theory. As such, the analysis is undertaken with the allowed presence of ignorance throughout the necessary operations. One feature of the analysis of US banks on their BFSR, is the impact of missing values in the financial characteristic values describing them. Using NCaRBS, unlike other traditional techniques, there is no need to externally manage their presence. Instead, they are viewed as contributing only ignorance. The comparative results shown on different versions of the US bank data set allows the impact to the presence of missing values to be clearly exposited. The use of the simplex plot method of visualizing data and analysis results furthers the elucidation possible with NCaRBS.

A deep understanding of different aspects of business performance and operations is necessary for a leading insurance company to maintain its position on the market and make further development. This chapter presents a clustering analysis for target group identification by locality, based on a case study in the motor insurance industry. Soft computing techniques have been applied to understand the business and customer patterns by clustering data sets sourced from policy transactions and policyholders’ profiles. Self organizing map clustering and k-means clustering are used to perform the segmentation tasks in this study. Such clustering analysis can also be employed as a predictive tool for other applications in the insurance industry, which are discussed in this chapter.

One of the most important aspects of operating a business is the forecasting of sales and allocation of resources to fulfill sales. Sales assessments are usually based on mental models that are not well defined, may be biased, and are difficult to refine and improve over time. Defining sales forecasting models for small- and medium-size business operations is especially difficult when the number of sales events is small but the revenue per sales event is large. This chapter reviews the challenges of sales forecasting in this environment and describes how incomplete and potentially suspect information can be used to produce more coherent and adaptable sales forecasts. It outlines an approach for developing sales forecasts based on estimated probability distributions of sales closures. These distributions are then combined with Monte Carlo methods to produce sales forecasts. Distribution estimates are adjusted over time, based on new developments in the sales opportunities. Furthermore, revenue from several types of sources can be combined in the forecast to cater for more complex business environments.

User relevance feedback plays an important role in the development of efficient and successful business strategies for several online domains such as: modeling user preferences for information retrieval, personalized recommender systems, automatic categorization of emails, online advertising, online auctions, etc. To achieve success, the business models should have some kind of interactive interface to receive user feedback and also a mechanism for user relevance feedback analyis to extract relevant information from large information repositories such as WWW. We present a rough set based discernibility approach to expand the user preferences by including the relevant conceptual terms extracted from the collection of documents rated by the users. In addition, a rough membership based ranking methodology is proposed to filter out the irrelevant documents retrieved from the information repositories, using an extended set of conceptual terms. This paper provides a detailed implementation of the proposed approach as well as its advantages in the context of user relevance feedback analysis based text information retrieval.

Supply chains form an integral cornerstone of the daily operations of today’s business enterprises. The effectiveness of the processes underlying the supply chain determine the steady-flow of raw material and finished products between entities in the marketplace. Electronic institutions facilitate free and fair competition between providers and suppliers vying for contracts from customers and manufacturers. We assume that contracts are awarded via competitive auction-based protocols. We believe that the efficiency of the scheduling and pricing strategies of the suppliers play an important role in their profitability in such a competitive supply chain. The suppliers can decide their scheduling strategy for completing the contracted tasks depending on their capacity, the nature of the contracts, the profit margins and other commitments and expectations about future contracts. Such decision mechanisms can incorporate task features including length of the task, priority of the task, scheduling windows, estimated future load, and profit margins. Robust, opportunistic task scheduling strategies can significantly improve the competitiveness of suppliers by identifying market niches and strategically positioning available resources to exploit such opportunities. Effective price adjustment mechanisms are also required to maximally exploit such market opportunities.

The need for soft computing technologies to facilitate effective automated tutoring is pervasive – from machine learning techniques to predict content significance and generate appropriate questions, to interpretation of noisy spoken responses and statistical assessment of the response quality, through user modeling and determining how best to respond to the learner in order to optimize learning gains. This chapter focuses primarily on the domain-independent semantic analysis of learner responses, reviewing prior work in intelligent tutoring systems and educational assessment. We present a new framework for assessing the semantics of learner responses and the results of our initial implementation of a machine learning approach based on this framework.

This paper presents a diagnosis system that helps the dentists to decide the course of treatment for dental caries. The inference mechanism of the system is based on the Bayesian Network (BN) and is designed to decide among various possible treatment plans. The system has been evaluated with the help of 13 different dentists to test its operational effectiveness. The system improves the confidence level of a dentist while deciding the treatment plan. As a result, it improves the effectiveness of the dentist and his/her business. Using this system, patients can also get information regarding the nature of treatment and the associated cost as well.

The complexity of current organization systems and the increase in importance of the realization of internal controls in firms makes the construction of models that automate and facilitate the work of the auditors crucial. A tool for the decision support process has been developed based on a multi-cbr system that incorporates two case-based reasoning systems and automates the business control process. The objective of the system is to facilitate the process of internal audit in small and medium firms. The multi-cbr system analyses the data that characterises each one of the activities carried out by the firm, then determines the state of each activity, calculates the associated risk, detects the erroneous processes, and generates recommendations to improve these processes. Therefore, the system is a useful tool for the internal auditor in order to make decisions based on the obtained risk. Each one of the case-based reasoning systems that integrates the multi-agent system uses a different problem solving method in each step of the reasoning cycle: a Maximum Likelihood Hebbian learning-based method that automates the organization of cases and the retrieval phase, an Radial Based Function neural network and a multi-criterion discreet method during the reuse phase and a rule based system for recommendation generation. The multi-cbr system has been tested in 14 small and medium size companies during the last 26 months in the textile sector, which are located in the northwest of Spain. The achieved results have been very satisfactory

In this study, an empirical mode decomposition (EMD) based neural network ensemble learning model is proposed for world crude oil spot price modeling and forecasting. For this purpose, the original crude oil spot price series were first decomposed into a finite and often small number of intrinsic mode functions (IMFs). Then the three-layer feed-forward neural network (FNN) model was used to model each extracted IMFs so that the tendencies of these IMFs can be accurately predicted. Finally, the prediction results of each IMFs are combined with an adaptive linear neural network (ALNN) to formulate a ensemble output for the original oil series. For verification, two main crude oil price series, West Texas Intermediate (WTI) crude oil spot price and Brent crude oil spot price are used to test the effectiveness of this proposed neural network ensemble methodology.