Background:
Research has shown various influential factors, but a standard for different treatments has not yet been developed. We want to identify principle predictors in IVF/ICSI treatment success rate using a machine learning approach.
Methods:
We selected the well-known filter, embedded and wrapper methods for selecting features and applied them to the random forest model. In the first step, the best reduce dimension method is selected using hesitant fuzzy sets (HFS). Then, the principle predictors are obtained by the proposed hybrid method. Finally, we evaluated the performance of the proposed method using MCC, runtime, Accuracy, and AUC, PPV, recall, and F-Score.
Results:
The proposed hybrid feature selection method has the best performance with obtaining the highest ACC (0.795), AUC (0.72), and F-Score (0.8) and lowest number of features (n = 7). Finally, the features selected are included FSH, 16Cells, FAge, Oocytes, GIII, Compact, and Unsuccessful.
Conclusion:
In the decision-making to select impressive features on success rate of infertility, HFS are suggested for the first time in our proposed method and trusted by multi-center dataset. The HFS utilized the standard deviation among of various criteria which could increase the quality of feature selection and reduce features numbers. The mean of pregnant and non-pregnant groups was significantly different in the Features selected, including FSH, FAge, 16Cells, Oocytes, GIII, and Compact. According to the results, FAge significantly correlated to FHR and CPR, and the highest FSH with 31.87% is obtained for FSH dose in the range of 10 to 13 (mIU/ml).