Investigation of Interdependence Issues among Multiple Responses in Multi Label Naïve Bayes

Abstract

Each instance in multi labeled data is associated with more than one response variable. Due to huge labels, association among labels should be preserved even after processing. Many traditional single label data mining classification algorithms have been adopted to handle multi label data. Among them, our research focuses on Multi Label Naïve Bayes algorithm. The main weakness of this algorithm is the lack of label dependency. While making prediction for the test instance, the Multi Label Naïve Bayes algorithm fails to preserve the correct label correlation among labels as in the original data set. This analysis would be helpful for the further improvement of this algorithm. This paper explores this weakness of the Multi Label Naïve Bayes algorithm. This problem is exhibited using different multi label data.