On the risk of extracting relevant information from random data.

Abstract

This comment constitutes a re-assessment of a recent study in which near-infrared spectroscopy (NIRS) was used to decode decision making. In the original study, the process of feature selection was carried out on all of the data, and those features which displayed the greater classification accuracy were selected, but no independent assessment or validation of the result was performed on a separated set of trials. In order to show the risk of this procedure, the same methodology was applied here to a set of random and independent time series instead of actual NIRS signals. This simulation produced statistically similar results to the original experimental study. It is my opinion that, from the reported classification accuracy of the original paper, no relevant or useful information is really obtained.