br Running time for the best solution br Feature

6.4.4. Running time for the best solution
Feature selection approaches can decrease the dimensional of feature space and reduce the computational complexity. Furthermore, it BQ-788 sodium salt can reduce the running time for the classification models. As can be observed in Figs. 8 and 9, we can obviously see that the IGSAGAW approaches achieved the best results followed by GAW approaches, followed by the baseline approaches. The main reason is that in IGSAGAW approaches, we firstly ranking the features according to its importance, which can increase the computational efficiency and decrease the difficulty for searching. From Figs. 2–9, it is seen that the IGSAGAW and GAW approaches have performed better than the baseline methods. The results of 10-fold cross validation on different models for WBC and WDBC data sets are concluded in Tables 9 and 10. The best and the optimum results are printed in bold. As can be observed in Tables 9 and 10, the results obviously indicate that our proposed model can achieve better performances than other comparative models.
7. Discussion
In this section, we will provide a discussion on the performance of different components of our proposed model. The proposed model is a hybrid intelligent classification method which fusion feature selection and classification. As previous mentioned, in order
The results of 10-fold cross verification based on WBC data set.
Underlying classifier
Accuracy Misclassification cost G-mean Running time
Note:
a Represents the best results, but is not the optimum results. b Represents the optimum results.
to straightforward assess the effectiveness of our proposal, we carried out a series of comparative experiments.
First of all, in order to verify the effect of our proposed feature selection approach, we first ran experiments on baseline classifiers with all features before applying GAW and IGSAGAW approaches. And the results can be seen in Figs. 2–9, from the results we can indicate that the IGSAGAW approach achieved the best results followed by GAW method, followed by baseline classifiers. To strengthen the advantageous of feature selection methods, we also investigate the CPU running time of our proposed model, the results as presented in Figs. 8 and 9, from this two figures, we can clearly see that our proposed model can decrease the computational efficiency for the breast cancer diagnosis.
Moreover, in order to verify the efficiency of the CSSVM classification algorithm proposed in this paper, we compare Polyadenylation with BP neural network and 3-NN. The experimental result are shown in Figs. 8 and 9, From the result shown in Fig. 2, the classification accuracy of IGSAGAW + BP neural network model achieved the best results for the WDBC and WBC data sets, which are 97.5%, 96.3%, followed by IGSAGAW + CSSVM model, which are 95.7%, 95.8%, the details are presented in Tables 9 and 10, where in Tables 9 and 10, we can clearly see that the BP neural network approaches achieved the best results, which we marked in bold. At the same time, from Fig. 5 we can clearly see that the hybrid model of IGSAGAW + BP also obtained the best results in terms of misclassification cost for the WDBC data set, which is 0.151, followed by IGSAGAW + CSSVM model, which is 0.202. Moreover, as for the performance of G-mean, the result of hybrid model of IGSAGAW + BP is 0.974, which is better than our proposed is 0.936 for the WDBC data set. From the above analysis, it can be seen that the model of IGSAGAW + BP can achieve better results than our proposed model, which in terms of the performances of classification accuracy, misclassification cost and G-mean. However, the most important is that the running time of our proposed model is the minimum, which is far less than IGSAGAW + BP hybrid model. Therefore from the above analysis, it can be seen that our proposed model is the most suitable method; it produces an excellent performances and only requires a moderate computational cost for solving breast cancer classification problems.