Table 1.
Variables Available for Analysis.
Table 2.
Nine algorithms used in the construction of prediction models.
Figure 1.
The optimization and prediction system.
SEER dataset prediction result A represents the 9*2 predictive results trained by nine data mining algorithms together with two variable selection methods and tested on the SEER testing dataset with all 20 variables. SEER prediction result B represents the 9*2 predictive results tested on the SEER testing dataset with 14 variables supported by both SEER and CMU-SO datasets. CMU-SO prediction result represents the 9*2 predictive results tested on the CMU-SO testing dataset with 14 variables supported by both SEER and CMU-SO datasets.
Figure 2.
The ROC curve from two different testing datasets.
A. Comparison of the predictive accuracy of three prognostic models: ANFIS together with GA, NB together with BSFS and the AJCC 7th TNM staging system using SEER testing dataset with 14 variables as a testing dataset. B. Comparison of the predictive accuracy of three prognostic models: LR together with BSFS, ANFIS together with GA and the AJCC 7th TNM staging system using the CMU-SO testing dataset as a testing dataset.
Table 3.
AUCa calculated by testing prediction models on SEERb.
Table 4.
AUCa calculated by testing prediction models on CMU-SOb.
Figure 3.
Survival rates at eight risk levels.
Comparative result between predictive survival rates to the real-world survival rates at eight different risk levels. The predictive survival rate is based on a predictive model built by LR together with BSFS.