MC3: A Multi-class Consensus Classification Framework

In this paper, we propose MC3, an ensemble framework for multi-class classification. MC3 is built on "consensus learning", a novel learning paradigm where each individual base classifier keeps on improving its classification by exploiting the outcomes obtained from other classifiers until a consensus is reached. Based on this idea, we propose two algorithms, MC3-R and MC3-S that make different trade-offs between quality and runtime. We conduct rigorous experiments comparing MC3-R and MC3-S with 12 baseline classifiers on 13 different datasets. Our algorithms perform as well or better than the best baseline classifier, achieving on average, a 5.56% performance improvement. Moreover, unlike existing baseline algorithms, our algorithms also improve the performance of individual base classifiers up to 10%. (The code is available at https://github.com/MC3-code.)