Incomplete data classification via positive approximation based rough subspaces ensemble

Abstract

Classifying incomplete data using ensemble techniques is a prevalent method for addressing missing values, where multiple classifiers are trained on diverse subsets of features. However, current ensemble-based methods overlook the redundancy within feature subsets, presenting challenges for training robust prediction models, because the redundant features can hinder the learning of the underlying rules in the data. In this paper, we propose a Reduct-Missing Pattern Fusion (RMPF) method to address the aforementioned limitation. It leverages both the advantages of rough set theory and the effectiveness of missing patterns in classifying incomplete data. RMPF employs a heuristic algorithm to generate a set of positive approximation-based attribute reducts. Subsequently, it integrates the missing patterns with these reducts through a fusion strategy to minimize data redundancy. Finally, the optimized subsets are utilized to train a group of base classifiers, and a selective prediction procedure is applied to produce the ensembled prediction results. Experimental results show that our method is superior to the compared state-of-the-art methods in both performance and robustness. Especially, our method obtains significant superiority in the scenarios of data with high missing rates.