This page is devoted to show the supplementary materials for the paper entitled "Cost-sensitive pattern-based classification for class imbalance problems" accepted in IEEE Access since April 25, 2019.

Classification results obtained by all the tested classifiers

In our experiments, we used cost matrices of size 2x2 where the main diagonal of the cost matrices was fixed as C(0; 0) = C(1; 1) = 0, the misclassification cost for each object of the majority class is C(0; 1) = 1, while for the misclassification cost for objects of the minority we use C(0; 1) = 2; 5; 10; 20 and the class imbalance ratio (IR) of the training database.

Actual minority Actual Majority
Predict Minority C(0, 0) = 0 C(0, 1) = 1
Predict Majority C(1, 0) = 2, 5, 10, 20, or IR C(1, 1) = 0
  • AUC results considering all the databases ->

Statistical Tests

Statistical results using Friedman's test (as a non-parametric test) and Finner's procedure (as a post-hoc procedure) [1]. The post hoc comparisons contain α = 0.05 and adjusted p-values.

  • Statistical results for all the tested contrast pattern-based classifiers; according to the AUC measure ->
  • Statistical results for all the tested classifiers not directly based on contrast patterns and our proposal based on contrast patterns; according to the AUC measure ->

CSPm+CACSP Implementation

CSPm+CACSP algorithm was implemented on .NET environment. In the file we provide the source code of the CSPm+CACSP algorithm and some examples using ARFF (WEKA database file) and COST (WEKA costmatrix file) files->


[1] J. Derrac, S. García, D. Molina, and F. Herrera, “A practical tutorial on the use of nonparametric statistical tests as a methodology for comparing evolutionary and swarm intelligence algorithms,” Swarm Evol. Comput., vol. 1, no. 1, pp. 3–18, 2011.