Classifying Words with 3-sort Automata

The process of grammatical inference involves acquiring a language or grammar through data analysis. This study explores various models aimed at deducing a non-deterministic finite automaton (NFA) with three types of states. These states are designed to accept certain words while rejecting others within a provided sample. The paper introduces a conversion method from the aforementioned three-state NFA to weighted-frequency and probabilistic NFAs. Subsequently, these probabilistic NFAs are employed in a classification task. Through experimental evaluation, our findings demonstrate the effectiveness of probabilistic NFAs in successfully tackling classification tasks across both real-world (The Waltz DB database [2] is utilized for the classification of peptides into two categories: amyloid, posing a potential threat, and non-amyloid, considered harmless) and artificial benchmark (languages generated by a regular expression) datasets.