Data Cleansing Based on Mathematic Morphology

In the field of bioinformatics and medical image understanding, data noise frequently occurs and deteriorates the classification performance, therefore an effective data cleansing mechanism in the training data is often regarded as one of the major steps in the real world inductive learning applications. In this paper the related work on dealing with data noise is firstly reviewed, and then based on the principle of mathematic morphology, the morphological data cleansing algorithms are proposed and two concrete morphological algorithms, DILATE-DC and CLOSE-DC, are realized. The experiments which are arranged on 15 UCI datasets show that these morphological data cleansing algorithms can effectively improve the classification performance, comparing with other relative methods.