Reviews: Efficient High-Order Interaction-Aware Feature Selection Based on Conditional Mutual Information

The problem of designing a feature selection algorithm capable of efficiently deal with high-order interaction among features is an interesting and open problem in the feature selection area. That is why this paper is appealing. However, there are several issues regarding the computational cost and the experimental setup that need a clarification in order to consider it for acceptance. It is said (lines 219-222) that "The described technique has been inspired by the intuition that probably two binary representatives of two different features interact on average better than two binary representatives of one feature"; however, no references or examples are provided to support this idea. On the other hand, when comparing the computational cost between the algorithm with and without binary representations (lines 215-219), the same values for t and s are considered. This is not a fair comparison as both cases are not taking the same level of information.