Prediction of compressive strength class of concrete with dominance based rough set approach

Abstract

Dominance based rough set approach is important in studies conducted with datasets con-taining uncertainty. In this study, a dataset consisting of 1030 samples obtained in the labora-tory regarding compressive strength of concrete has been considered. The decision attribute, which has continuous values, has been made discrete for applying dominance relation. In order to measure performance, samples in the dataset have been divided into two groups: the training set and the testing set. This process has been done in a way that corresponds to the distribution of each class within the dataset. On the other hand, since there is a class which has more or less samples than the others, synthetic data generation has been done with Synthetic Minority Oversampling Technique (SMOTE) in order to handle the between-class imbalance problem and equalize the number of samples in the classes. As a result, the training set has been made perfectly balanced. A decision-support model which extracts if... then... exact decision rules has been designed to be used in determining the quality or compressive strength of the concrete samples by using dominance based rough set approach. Performance of these rules on the testing set through the confusion matrix has been discussed. The exper-imental results show that performance of the exact decision rules induced by the dominance rough set approach on the testing set is significant.