Comparison of Shapley-Shubik and Banzhaf-Coleman power indices applied to aggregation of predictions obtained based on dispersed data by k-nearest neighbors classifiers
Małgorzata Przybyła‐Kasperek, Filip Smyczek · Procedia Computer Science · 2022
In this paper a new method of fusion predictions obtained based on dispersed data is proposed. In the method a power index is used. This approach allows to calculate the real power of prediction vectors generated based on local data with using the k-nearest neighbors classifier. The use of two power indices: Shapley-Shubik and Banzhaf-Coleman power index is analyzed. The influence of k-parameter value and the value of quota in simple game on the classification accuracy is also studied. The obtained results are compared with the approach in which the power index was not used. It was found that the proposed method of using the power index improves the classification accuracy. Moreover, both analyzed power indices generate comparable results.