Table 2.

Record-wise and subject-wise cross-validation (CV) techniques

CV group	CV technique	Description
Record-wise group	Stratified k-folds CV (skfcv)	With this technique, the dataset is divided into k blocks (folds) in a stratified manner [12]. One of the k blocks is selected as the validation set, while the remaining k–1 blocks constitute the training set. This process is repeated k times, with k = 10. As we are dealing with a binary classification problem, stratification is essential to ensure an equal distribution of both classes (persons with Parkinson’s disease and healthy controls) in each fold.
	Leave-one-out CV (loocv)	In this technique, only one record is left out for each learning process [12]. We consider n the number of records of our dataset, training is done on n-1 records, and validation is done on a single record. This process is repeated n times.
	Repeated stratified k-folds CV (rskfcv)	This technique is similar to stratified k-folds CV, but it is repeated n times [12]. We consider n the number of repetitions and k the number of blocks. This process is repeated k × n times, with k = 10 and n = 5. This method guarantees a more accurate estimate than without repetition.
Subject-wise group	Stratified-group k-folds CV (sgkfcv)	Using this technique, the dataset is divided into k blocks in a stratified manner with group of subjects [13]. This means that if a subject with a set of records is in block k, the recordings of that person do not occur in block k–1. One of the k blocks is chosen as the validation set, while the remaining k–1 blocks constitute the training set. This process is repeated k times, with k = 10.
	Leave-one-group-out CV (logocv)	In this technique, we leave out the records of only one group of subjects for each learning process [12]. We consider g the number of people in our dataset, learning is done on g–1 groups, and validation is done on a single group. This process is repeated g times.
	Repeated stratified-group k-folds CV (rsgkfcv)	This technique is similar to stratified-group-k-folds CV, but it is repeated n times [13]. We consider n the number of repetitions and k the number of blocks. This process is repeated k × n times, with k = 10, and n = 5. This method guarantees a more accurate estimate than without a repetition.