Maykon Marcos Júnior – Universidade Federal de Santa Catarina – https://orcid.org/0009-0002-5432-8995
Guilherme de Brito Santos – Universidade Federal de Santa Catarina – https://orcid.org/0009-0005-8748-4814
João Gabriel Mohr- Universidade Federal de Santa Catarina – https://orcid.org/0009-0004-5300-7940
Andressa Silveira Viana Maurmann- Universidade Federal de Santa Catarina – https://orcid.org/0009-0004-7948-6708
Luísa Bollmann- Universidade Federal de Santa Catarina – https://orcid.org/0009-0005-7285-7905
Arthur Machado Capaverde – Universidade Federal de Santa Catarina – https://orcid.org/0009-0002-0544-9934
Cristian Alexandre Alchini- Universidade Federal de Santa Catarina – https://orcid.org/0009-0004-2510-7338
Maite Fortes Vieira- Universidade Federal de Santa Catarina – https://orcid.org/0009-0007-4229-5483
Lucas de Castro Rodrigues Pereira- Universidade Federal de Santa Catarina – https://orcid.org/0009-0002-1510-3021
Isabela Cristina Sabo- Universidade Federal de Santa Catarina – https://orcid.org/0000-0003-4246-3997
Aires José Rover- Universidade Federal de Santa Catarina – https://orcid.org/0000-0003-1070-5357
This article presents the results of the Concil-IA Project, an interdisciplinary initiative developed at the Federal University of Santa Catarina (UFSC) in partnership with the institution’s Small Claims Court, aimed at creating an Artificial Intelligence (AI)-based Online Dispute Resolution (ODR) platform for consumer conflicts. The research consolidated its contributions into three main areas: predictive modeling, explainability, and validation of the digital interface. In the field of predictive modeling, a regression model was developed using 1,851 anonymized judicial decisions related to air transport disputes. Several Machine Learning techniques (Decision Tree, Random Forest, and AdaBoost) were tested, and the Decision Tree Regressor was selected for balancing performance, interpretability, and computational efficiency. The model achieved satisfactory results, with a mean absolute error of approximately R$ 1,672 and a root mean squared error of R$ 2,286, confirming its suitability for real-world conciliation scenarios. Regarding explainability, Explainable Artificial Intelligence (XAI) methods were applied through SHAP (SHapley Additive exPlanations). This approach enabled both global interpretation of the model and local explanations for specific cases, highlighting as the most relevant factors in compensation predictions the flight delay or cancellation and the absence of assistance provided by the airline. Finally, the model was incorporated into a responsive web platform (concilia.ufsc.br / app.concilia.ufsc.br), developed in WordPress and validated with conciliators and court staff. Usability tests, carried out through structured questionnaires, showed high levels of acceptance, with average scores between 4.8 and 5.0 in clarity, organization, comfort, and perceived effectiveness. The findings confirm the technical feasibility and institutional relevance of Concil-IA as a digital innovation for conflict resolution, particularly judicial conciliation. Future steps include expanding the dataset, applying the platform to other areas of law, and strengthening its interoperability with national digital justice policies.
