Evaluation of the expert system shells EX-TRAN and TIMM as rule-building tools for classification purposes

Abstract This article describes results obtained with two rule-building expert systems using simulated data sets. It was found that certain data structures (e.g. strongly correlated data or data containing outliers) cannot be handled in an optimal way. In some cases (e.g. with strongly overlapping data) one can observe from the rule set that the classification is not robust. This requires expertise however. Our results lead us to conclude that these expert systems do not yet possess sufficient expertise to be used without problems by non-experts or in conjunction with analytical instruments.