Using misclassification data to improve classification performance
Ratchakoon Pruengkarn, Chun Che Fung, Kok Wai Wong · 2015
Improvement of classification accuracy is importance in data analysis problems. Enhancement of techniques have been proposed previously to address the problems as regard to classification performance, however, the issues of misclassification and noise elimination in the early stage of processing have been ignored by many researchers. If these problems were addressed, the performance of the classification may be improved. In this paper, a framework for misclassification analysis is proposed. Feature selection using Fuzzy C-means can be implemented in the early stage of model building. Then, ensemble techniques using majority vote algorithm could be incorporated in order to reduce misclassification. The proposed technique has shown an improved classification performance in terms of accuracy rate. The performance was improved for both cases of binary and multiclass data sets at 14.36% on average. In addition, the performance of the classification model for multiclass data sets improved more in comparison to the binary data sets.