Retrieving meaningful information from high dimensional dataset is an important and challenging task. Normally, medical dataset suffers from several issues such as curse of dimensionality problem, uncertainty, presence of missing values, non-relevant and redundant attributes, etc. Any machine learning technique applied on such data (without any preprocessing) by and large takes a considerable amount of computational time and may degrade the performance of the model.

In this article, R-Ensembler, a parameter free greedy ensemble attribute selection method is proposed adopting the concept of rough set theory by using the attribute-class, attribute-significance and attribute-attribute relevance measures to select a subset of attributes which are most relevant, significant and non-redundant from a pool of different attribute subsets in order to predict the presence or absence of different diseases in medical dataset. The main role of the proposed ensembler is to combine multiple subsets of attributes produced by different rough set filters and to produce an optimal subset of attributes for subsequent classification task. A novel n number of set intersection method is also proposed to reduce the biasness during the time of attribute selection process. Before selecting the minimal attribute set from a given data by the proposed R-Ensembler method, the dataset is preprocessed by the k nearest neighbour (kNN) imputation method for missing value treatment.

Experiments are carried out on seven benchmark medical datasets collected from University of California at Irvine (UCI) repository. The performance of the proposed ensemble method is compared with five state-of-the-art attribute selection algorithms, results of which are measured using three benchmark classifiers viz., Naïve Bayes, decision trees and random forest. Experimental results clearly justify the superiority of the proposed R-Ensembler method over other attribute selection algorithms. Results of paired t-test performed on average accuracies produced by different classifiers simulated on the reduced data sets achieved by the proposed and counter part attribute selection methods confirm the statistical significance of the better reduced attribute subsets achieved by the proposed R-Ensembler method compared to others.

The proposed ensemble method turned out to be very effective for selecting high relevant, high significant and less redundant attributes from a pool of different subsets of attributes.