Effective Feature Selection Methods for Vaginal Birth After Cesarean Data

This research aims to build an effective feature selection method to predict the success of Vaginal Birth After Cesareanean (VBAC) based on medical record data. In automatic data processing using a machine learning approach, feature selection is useful for reducing data dimensions, ultimately reducing algorithm complexity and speeding up processing time. In the medical aspect, feature selection from medical record data contributes to determining significant predisposing factors in determining the success of VBAC. The selected features can be used as a simpler reference by medical personnel, making retrieving medical record data shorter and more efficient. We propose a feature selection method by combining PSO and PCA because PSO is claimed to provide optimal feature selection results. However, the user cannot determine the final number of features selected. While PCA allows users to determine the dimension-reduced data, the reduced features do not correspond to the original data features. Logistic Regression (LR) was used as a classifier. Experimental results show that the combination of PSO-PCA provides the best results in prediction model performance and number of features compared to original data without a feature selection process. PSO reduces 42 features to 37, then PCA is combined to reduce dimensions, with size 8 giving the most optimal results, with PCC 90.0% and AUC 0.832. This result indicates that our proposed methods can select significant predisposing factors well in the case of VBAC and reduce the dimensions of the input data for the classifier while still providing good prediction performance.