Search

SMART Software for Decision Makers (SSDM) was a Department of Trade and Industry (DTI) initiative running during the period 1998-2000. The Centre for Adaptive Systems at the University of Sunderland was appointed by the DTI to run one of the two Demonstrator Clubs in the UK. The purpose of these clubs was to facilitate technology transfer between academia and industry, in the areas of fault diagnosis and prediction, intelligent control systems, and knowledge discovery in databases (KDD). Sunderland SSDM Club decided to use the various industrial members problems and data to build a number of “mini-demonstrators”. In the KDD cluster, three demonstrator applications were developed, accompanied by supporting material and a series of seminars, which illustrated the various stages in the KDD process to all club members. This paper describes three KDD application demonstrators, developed with data from a manufacturing company, with consultants in business clustering, and from data from a local police force, to investigate the phenomena of repeat victimization.The work involved data preprocessing, data transformation, data mining, and the development of visual tools for interpretation. With both the business clustering and police data much of the time was spent in data preparation, and so tools were developed so that the members could conduct their own data mining and interpretation experiments, lessening the need for the extraction of domain knowledge from the members.

Comprehens ibility is an important factor in medical predictive modelling as it dictates the credibility and even acceptability of a model. Generally, the performance of a model has always been the primary focus in most data mining jobs. Where there are serious risks posed by the decisions made by a model, it is not feasible to view comprehensibility aspects of a model as secondary to performance. While model comprehensibility is a topic that has aroused a lot of interest with two conference workshops (AI-UCAI'95 & AAAI 2005) placing it as its keynote issue and many papers written about it, there are no empirical methods of measuring it or even one consistent way to define it. It is generally accepted that smaller models are more comprehensible than larger ones. This forms the basis of most researches conducted in this area. In this paper, we investigate the efficacy of using multiobjective optimization in the Pareto sense to meet comprehensibility demands of models. Some of the objective functions used in this paper are novel while others have been used in other researches before. The results obtained show that incorporating aspects of comprehensibility in the induction process models does not necessarily retard the performance of models and could actually improve the performance versus complexity trade-off of evolutionary polynomial-fuzzy structures

Decision tree induction has been studied extensively in machine learning as a solution for classification problems. The way the linear decision trees partition the search space is found to be comprehensible and hence appealing to data modelers. Comprehensibility is an important aspect of models used in medical data mining as it determines model credibility and even acceptability. In the practical sense though, inordinately long decision trees compounded by replication problems detracts from comprehensibility. This demerit can be partially attributed to their rigid structure that is unable to handle complex non-linear or/and continuous data. To address this issue we introduce a novel hybrid multivariate decision tree composed of polynomial, fuzzy and decision tree structures. The polynomial nature of these multivariate trees enable them to perform well in non-linear territory while the fuzzy members are used to squash continuous variables. By trading-off comprehensibility and performance using a multi-objective genetic programming optimization algorithm, we can induce polynomial-fuzzy decision trees (PFDT) that are smaller, more compact and of better performance than their linear decision tree (LDT) counterparts. In this paper we discuss the structural differences between PFDT and LDT (C4.5) and compare the size and performance of their models using medical data