One-RM: An Improved One-Rule Classifier
Abdur Rehman Mahmood, Wei Lei · Bangladesh Journal of Scientific and Industrial Research · 1970
One-R algorithm is a simple algorithm which exhibits quite good predictive accuracy for a large class of data. When compared to the more complex algorithms having better predictive accuracy, One-R provides the baseline accuracy for testing new machine learning algorithms. However, the simplicity of One-R means that it has there is a compromise between accuracy and complexity. Often, the accuracy of One- R can be further increased without making it significantly complex. The resulting algorithm as proposed in this paper, One-RM performs equal to One-R in most of the cases and sometimes outperforms One-R by significant margin. Theoretical analysis suggests that One-RM used in conjunction with One-R always performs either better or equal to One-R. Experimental analysis shows that One-RM is a viable alternative to One-R when used as a separate classification rule. Key words: One-RM, One-R algorithm, Algorithm, Accuracy and Complexity. DOI: 10.3329/bjsir.v44i2.3668 Bangladesh J. Sci. Ind. Res. 44(2), 171-180, 2009