A statistical comparison of feature selection methods is performed. Feature selection is an important issue in Data Mining and Data Science, and a comparison of the results obtained from different methods is hard to be performed. Then, the evaluation of metrics and ways of comparisons is an important matter of study. Our study is performed on a real dataset previously analyzed in the literature containing a small number of records, drawing the attention on the conclusions to be applied where poor statistical confidence levels of significance can be obtained because of a relative low number of samples are present. The use of inter rater agreement coefficients is introduced as a novel approach extending a previous study. Boruta and tree-based methodologies perform rather well even in small data as it is shown. Our metrics can be used to guide the expert opinion in order to take the final decision. This work extends the results obtained in a previous analysis performed on the mentioned dataset.