Rough-Fuzzy approach to Attribute Importance and Attribute Ranking

Attribute importance refers to a vector of weights that are assigned to attributes and
describe the effect on the overall performance or outcome of a model, system, or process.
This concept is commonly used in a variety of fields, including machine learning, statistics,
data analysis, and decision making.

In machine learning, the importance of attributes helps determine which features have the
greatest impact on model predictions. Attribute importance and ranking can have several
benefits including feature selection, model interpretability and dimensional reduction.
Attribute importance can be calculated by different techniques including Random Forest
Feature Ranking (similar to Boruta) and Permutation Feature Importance.
Using the Random forest algorithm, the feature importance can be measured as the average
impurity decrease computed from all decision trees in the forest. The second method focuses
on observing how predictions of the ML model change when we change the order of variables.

In thí talk, the RAFAR (Rough-fuzzy Algorithm For Attribute Ranking) methodology will be presented.
This is a hybrid approach that combines the discernibility relation of the rough set theory and the
ranking method from intuitionistic fuzzy (Atanassov-IFS) theory. The RAFAR methodology consists
of two main steps:
(1) construction of a fuzzy pairwise comparison matrix (called Intuitionistic Fuzzy Preference
Relation (IFPR)) for the set of attributes, and
(2) converting this matrix into the optimal consistent weight vector,

Both theoretical foundation as well as the experimental results will be presented and discussed
during the talk.