Evolutionary Fuzzy Rules for Ordinal Binary Classification with Monotonicity Constraints

We present an approach to learn fuzzy binary decision rules from ordinal temporal data where the task is to classify every instance at each point in time. We assume that one class is preferred to the other, e.g. the undesirable class must not be misclassified. Hence it is appealing to use the Variable Consistency Dominance-based Rough Set Approach (VC-DRSA) to exploit preference information about the problem. In this framework, the VC-DomLEM algorithm has been used to generate the minimal set of consistent rules. Every attribute is then fuzzified by first applying a crisp clustering to the rules’ antecedent thresholds and second using the cluster centroids as indicator for the overlap of neighboring trapezoidal normal membership functions. The widths of the neighboring fuzzy sets are finally tuned by an evolutionary algorithm trying to minimize the specificity of the current fuzzy rule base.