Identificação de perfis de usuários nos aeroportos no Brasil usando técnicas não supervisionadas de aprendizado de máquina

Abstract

The growth of the airport sector in Brazil has increased competition among major airports, making it essential to understand the factors that influence passenger experience. This study aimed to identify user profiles at the 20 busiest airports in the country using unsupervised machine learning techniques on a dataset from the National Civil Aviation Secretariat. Clustering methods, such as K-Means and Hierarchical Clustering, combined with Multiple Correspondence Analysis (MCA), were applied to segment passengers based on behavioral and satisfaction characteristics. Cluster definition was based on the silhouette coefficient, resulting in three distinct groups for the K-Means method and two for the hierarchical method. The attributes considered included age, education level, household income, and satisfaction variables such as comfort, cleanliness, availability of power outlets, and restroom quality. The identified profiles were primarily defined by overall passenger satisfaction, with an emphasis on factors such as comfort, cleanliness, and ease of movement within the airport. The analysis highlighted that user experience-related variables were predominant in the formation of clusters, allowing for targeted actions aimed at personalizing and improving services at Brazilian airports.