Result dataset for our experimental analysis on multi-cepstral projection representation strategies for dysphonia detection

Database containing the results of the analyzed versions of the framework proposed in our paper submitted to the journal Sensors (Basel) under the title "An experimental analysis on multi-cepstral projection representation strategies for dysphonia detection".

In this database, we have the following information:

"gender": Gender of individuals referring to the selected voice database. For this field, we have the following possible values: “male” for a selection of male individuals, “female” for a selection of female individuals, and “both” for a selection considering both genders.

“Techniques”: Concerns about the techniques for extracting cepstral coefficients that we are analyzing. The identifier “nonceps” refers to the use of non-cepstral features.

“vowel”: Vowel considered in the database selection. The following values are possible: “a”, “i” and “u”.

“intonation”: Tone used by individuals when pronouncing the analyzed vowel. Possible values are: “h” for high; “l” for low; “n” is normal; and “lhl” for low-high-low.

“coordinates”: Number of coordinates that make up the feature vector that represents the voice signal after the dimensionality reduction routines.

“scale”: Normalization function used on the feature vector. The possible values of this field are the following: “MinMax” for the min-max scale; “Robust” for the robust scale; “Standard” for the standard scale; and “Unscaled” for the unscaled vector.

“ACC”: Accuracy obtained by the analyzed version on the considered voice database clipping.

“AUC”: Area under the ROC curve obtained by the analyzed version on the considered voice database clipping.

“EER”: Equal Error Rate obtained by the analyzed version on the considered voice database clipping.

“F1”: F1-score obtained by the analyzed version on the considered voice database clipping.

“EH”: Rate of healthy voice signals classified as pathological on the considered voice database clipping.

“EP”: Rate of pathological voice signals classified as healthy on the considered voice database clipping.

“KFCV”: Average accuracy score of a 5-fold Cross Validation over the training dataset on the considered voice database clipping.

“Balancing”: Indication of the use of balancing technique (SMOTE) by the considered framework version.

“Classifier”: Classifier used, being possible the use of Random Forest (RF), Logistic Regression (LR), and Support Vector Machine (SVM).

“Multi-Projection”: Multi-projection strategies employed by the evaluated technique.

“Features”: Type of feature that defines the feature vector. In this case, the following values are possible in this field: “NonCeps” for non-cepstral features; “Ceps” for cepstral features only; and “Ceps and NonCeps” for features of cepstral and non-cepstral types.
.

It is worth noting that the symbol “-”, present in some fields, represents the “non-use” of any technique of the type indicated by the field. For example, in the case of the “Balancing” field, the value “-” means that no data balancing technique was used in the evaluated version of the framework.