[Paper Review] Aggregation of Classifiers: A Justifiable Information Granularity Approach

Tien Thanh Nguyen, Xuan Cuong Pham+2 more|arXiv (Cornell University)|Mar 15, 2017

Rough Sets and Fuzzy Logic34 references4 citations

TL;DR

This paper proposes a novel ensemble learning method that uses interval-valued information granules to aggregate heterogeneous classifiers, modeling uncertainty and diversity through bounds of interval memberships rather than fixed numeric values. The approach improves classification performance across UCI datasets, outperforming benchmark methods including AdaBoost, Bagging, and multiple fixed/learned combining rules by leveraging granular uncertainty in decision-making.

ABSTRACT

In this study, we introduce a new approach to combine multi-classifiers in an ensemble system. Instead of using numeric membership values encountered in fixed combining rules, we construct interval membership values associated with each class prediction at the level of meta-data of observation by using concepts of information granules. In the proposed method, uncertainty (diversity) of findings produced by the base classifiers is quantified by interval-based information granules. The discriminative decision model is generated by considering both the bounds and the length of the obtained intervals. We select ten and then fifteen learning algorithms to build a heterogeneous ensemble system and then conducted the experiment on a number of UCI datasets. The experimental results demonstrate that the proposed approach performs better than the benchmark algorithms including six fixed combining methods, one trainable combining method, AdaBoost, Bagging, and Random Subspace.

Motivation & Objective

• To address the limitations of fixed combining rules in ensemble learning that ignore uncertainty and diversity among base classifiers.
• To model classifier uncertainty and diversity using information granules represented as intervals, rather than crisp membership values.
• To develop a discriminative decision model that considers both interval bounds and interval length for improved classification accuracy.
• To evaluate the proposed method on heterogeneous ensemble systems using diverse learning algorithms across standard UCI datasets.
• To demonstrate superior performance compared to established ensemble methods such as AdaBoost, Bagging, and Random Subspace.

Proposed method

• Construct interval membership values for each class prediction using information granule theory, representing uncertainty and diversity from base classifiers.
• Use the lower and upper bounds of intervals as well as interval length as features in a discriminative decision model.
• Apply a meta-learning approach where the final decision is derived from interval-based granular representations rather than numeric scores.
• Select 10 and then 15 diverse learning algorithms to form heterogeneous ensemble systems for robust evaluation.
• Train and test the ensemble on multiple UCI benchmark datasets to assess generalization and performance.
• Compare the interval-based aggregation with six fixed combining rules, one trainable method, AdaBoost, Bagging, and Random Subspace.

Experimental results

Research questions

• RQ1 Can interval-valued information granules effectively represent uncertainty and diversity among base classifiers in ensemble systems?
• RQ2 Does incorporating interval bounds and interval length into the decision model improve classification performance over traditional aggregation methods?
• RQ3 How does the proposed granular aggregation approach compare to established ensemble techniques like AdaBoost and Bagging across diverse datasets?
• RQ4 To what extent does the use of heterogeneous base learners enhance the robustness and accuracy of the interval-based ensemble?
• RQ5 Is the proposed method more justifiable in terms of uncertainty modeling compared to fixed or learned numeric combination rules?

Key findings

• The proposed interval-based aggregation method achieved superior classification accuracy compared to six fixed combining rules on the tested UCI datasets.
• The method outperformed the trainable combining method, indicating that granular uncertainty modeling provides a more effective basis for ensemble decisions than learned numeric weights.
• The heterogeneous ensemble using 15 diverse base learners consistently delivered higher performance than simpler ensemble configurations.
• The inclusion of interval length as a discriminative feature contributed significantly to improved decision boundaries and generalization.
• The approach demonstrated robustness across multiple datasets, with consistent performance gains over AdaBoost, Bagging, and Random Subspace.
• The results validate that information granules provide a more justifiable and interpretable framework for modeling classifier uncertainty in ensemble learning.