An Ensemble Classifier Based on kNN with an Interval Threshold Strategy

In this contribution, there is applied interval modelling for improving the quality of classification by the binary ensemble classifiers based on kNN in the case of microarrays. The proposed algorithm has a higher classification quality than the base classifying models. The algorithm is based on the interval-valued aggregation of intervals determined based on certainty coefficients of individual kNN classifiers where the thresholding strategy relies on interval obtained as an output of the interval-valued aggregation. As a result, there is considered the option of lack of decision (in some sense empty class label). The interval-valued aggregation is treated as a hyperparameter of the model so there is provided an analysis of the usefulness of the discussed interval-valued aggregation functions. The obtained results prove that in practical applications the newly proposed algorithm can be successfully used in the fields, such as medicine, where the option of “no decision” is acceptable and simultaneously the accuracy of the given decision is crucial.