An Automata-Based Feature Engineering Framework for Hepatitis Prognosis Modeling

Abstract

Hepatitis prognosis remains a challenging clinical task due to the complex and progressive nature of liver dysfunction, where patient outcomes are influenced by sequential changes in biochemical markers, complications, and treatment response. Most existing machine learning approaches rely on static feature a representation, which limit their ability to capture real-world disease progression and often reduces interpretability. To address this limitation, this study proposes an automata-based prognostic modeling framework that explicitly represents Hepatitis progression through deterministic state transitions aligned with clinical reasoning.In the proposed methodology, conventional clinical attributes are first pre-processed and transformed into symbolic representations, which are then processed using a deterministic finite automaton to model progression patterns. From the resulting state transitions, novel high-level features are extracted, capturing progression severity, transition dynamics, and response behavior. These automata-embedded features are combined with original clinical variables and evaluated using multiple machine learning classifiers on the Hepatitis dataset from the UCI Machine Learning Repository.Experimental results demonstrate that models incorporating automata-derived features consistently outperform conventional feature-based approaches across accuracy, error metrics, and stability analysis. In particular, high-performance classifiers and hybrid ensemble combinations achieve substantial gains in predictive accuracy, highlighting the effectiveness of progression-aware feature extraction. The proposed framework not only improves prognostic performance but also enhances interpretability, offering a clinically aligned and reliable decision-support approach for Hepatitis outcome prediction.