Improving classification with IF-THEN rules for multidimensional datasets

Abstract

The multidimensional datasets are becoming widespread in both scientific and business computing. Dealing efficiently with high-dimensional data is a challenge for researchers in the database field.

This paper proposes BIMA, a new classification method which uses the discovered rules in RIPPER classification in order to select the boundary instances of multidimensional datasets and to multiply them in the training phase of the next evaluation. In the testing phase, the instances were kept unchanged. In the experimental part it was demonstrated that the BIMA is a promising algorithm for improving the IF-THEN rules classification accuracy and also for improving the TP value of the multidimensional datasets classes. The efficiency of the proposed algorithm is proved by using the UAB graduates’ responses datasets.