Evaluation of Logistic Regression for Learning Preferences in Obtaining Weights for the Selection of Small Drones

The advancement of defense technologies has broadened the variety of remotely piloted aircraft, making the selection of the best alternative a multi-criteria decision problem. This study investigates a Preference Learning approach to estimate criterion weights through pairwise comparisons, aiming to support the selection of small drones in the Brazilian Marine Corps. A dataset was constructed from expert evaluations, and a logistic regression model was trained to learn, from these preferences, the relative contribution of the criteria. Five variables were analyzed: endurance, range, maximum speed, wind resistance, and weight. The results indicated high predictive accuracy (95.24%) and weights consistent with operational logic, with emphasis on maximum speed, range, and endurance. It is concluded that logistic regression can complement multi-criteria methods, increasing the efficiency in prioritizing alternatives.