A Supervised Learning Approach for the Fusion of Multiple Classifier Outputs
C. Ulas, B. Koroglu, C. Bekar, O. Burcak, and O. Agin
R&D and Special Projects Department, Yapi Kredi Bank, Kocaeli, Turkey
Abstract—In this paper, we propose an alternative method for the fusion of multiple classifier outputs to obtain a potential improvement in the classification performance. Our method mainly operates on decisions of several classifiers, and the individual classifiers, which are trained on different data sets, share common classes and some of them also individually deal with classifying different class labels. For this reason, we first aim to define a common space to represent all the class labels and then propose a supervised learning approach to effectively recognize patterns of the tuples that are formed by concatenating classifiers outputs based on top-N rankings. To evaluate the performance of our method, we utilize Random Forest (RF) classifier for this multiclass classification problem and the results demonstrate that our method can achieve promising performance improvement in true positive rate of classification compared to that of the best performing individual classifier yields.
Index Terms—classifier output fusion, random decision forest, supervised learning, decision support systems
Cite: C. Ulas, B. Koroglu, C. Bekar, O. Burcak, and O. Agin, "A Supervised Learning Approach for the Fusion of Multiple Classifier Outputs," International Journal of Signal Processing Systems, Vol. 4, No. 3, pp. 198-203, June 2016. doi: 10.18178/ijsps.4.3.198-203
Cite: C. Ulas, B. Koroglu, C. Bekar, O. Burcak, and O. Agin, "A Supervised Learning Approach for the Fusion of Multiple Classifier Outputs," International Journal of Signal Processing Systems, Vol. 4, No. 3, pp. 198-203, June 2016. doi: 10.18178/ijsps.4.3.198-203
Previous paper:Network Element Simulation Based on Log Files
Next paper:Vulnerability Research and Mapping of Campus Network
Next paper:Vulnerability Research and Mapping of Campus Network