Novel feature selection methods for construction of accurate epigenetic clocks
Fig 4
The workflow for feature selection and model evaluation.
Feature selection was performed on training data for each iteration of 10-fold cross validation. The selected features of each iteration are aggregated into a list for each feature selection method type. The unique selected features for each method are collected into a dataframe where post-selection processes such as intersections are performed. We add the results to a dataframe. Each column of selected features in the results dataframe (each representing a different feature selection method) is tested using another training-testing split on the original data. This is done 10 times for 10-CV with the average of all scores being the performance estimate for that feature selection method.