Improving LAIM discretization method for multi-label data using evolution strategy
Amene Esmaeili Rad, Hossein Nezamabadi–pour, Mahdi Eftekhari · 2018
Recently, multi-label learning has become challenging topic and has attracted many attention. Many classification algorithms require discrete data as input; therefore, discretization has an important role for data preprocessing. Furthermore, many other benefits such as improvement in algorithm's performance are achieved after data discretization. Many multi-label datasets have real-valued input variables, whilst multi-label discretization techniques have not been addressed appropriately, so far. Recently LAIM discretization method has been proposed for such data. In this paper, an improvement of this method is presented using (1+1)-ES, which is a simple evolution strategy (ES). The experimental results over 8 datasets confirm the effectiveness of the proposed method.