There are variables increasing vulnerability while others decrease it. What are the best pre-processing methods to account for this relationship to ensure adequate results? Is it always necessarily a max-normalization or another rescaling procedure? How can we interpret negative and positive signs in eigenvectors?