Comparative Analysis of Automata-Based and Machine Learning Approaches for Postoperative Risk Prediction in Thoracic Surgery

Lung cancer thoracic surgery is a particularly high-risk procedure in terms of both the occurrence of postoperative complications and mortality, which requires the implementation of decision-support systems that are correct, easily understood and interpreted. Conventional statistical and Machine Learning (ML) methods have been shown to be promising for predicting one-year survival using datasets such as the Thoracic Surgery Data in the UCI repository, which include preoperative variables such as age, comorbidities, pulmonary function tests, etc. Nevertheless, they are usually not accepted clinically as they are not transparent since they are black boxes. The automata theory, specifically Deterministic Finite Automata (DFA) and rule-based state machines, provides a complementary model of clinical pathways, with explicit state transitions and rule enforcement. The automata can perform powerful data validation, detect high-risk patterns, and perform real-time patient classification through deterministic traceability by directly encoding medical logic into transitions. This review comparatively focuses on automata-based methods and ML models and highlights the relative merits of their interpretability, computational efficiency, and adherence to protocols, versus their predictive accuracy and multi-signal combination. Moreover, it makes evident the opportunities for hybrid architectures with automata-driven capabilities to improve ML predictions, thereby balancing transparency and performance. Finally, automata integration into clinical decision support offers a way to reliable, explainable, and deployable risk prediction instruments in the field of thoracic surgery to aid in achieving better patient safety and improved surgeon decision-making.