A feature selection algorithm for online multi-label data
Subject Areas : Information Technology in Engineering Design (ITED) Journalmaryam rahmaninia 1 , sondos bahadori 2
1 - Department of Computer Engineering,, Qasreshirin Branch, Islamic Azad University, Qasreshirin,, Kermanshah Iran
2 - Department of Computer Engineering, Ilam Branch, Islamic Azad University, Ilam, Iran,
Keywords: Feature selection, Multi label dataset, data with feature stream, multi variate feature selection,
Abstract :
Currently, we are facing a large volume of multi-label training data, where each training instance has more than one label. One of the main challenges with this type of training data is the sudden increase in its size, which impacts the efficiency of data mining algorithms. Multi-label feature selection methods reduce the dimensions of this data by selecting a subset of important initial features. On the other hand, in today's world, data is gradually being added over time. Therefore, multi-label feature selection algorithms need to be used in an online fashion. Although several online multi-label feature selection algorithms have been proposed recently, none of them have considered the dependency between labels to determine the relationship between features and labels.To address this issue, this paper presents an online multi-label feature selection method that incorporates multiple mutual information measures into the feature selection process. Additionally, several experiments have been conducted on multiple multi-label feature selection algorithms and datasets to demonstrate the effectiveness of the proposed method. The results obtained indicate that the proposed method performs well in terms of various evaluation metrics.