Avtomatsko generiranje razlag pri diagnostiki shizofrenije z uporabo testa tekočnosti govora

Schizophrenia is a chronic and severe mental disorder affecting millions of people worldwide. To support the diagnostic process, this thesis presents the development of an automated pipeline for detecting schizophrenia based on audio recordings of verbal fluency tests from 126 Slovene-speaking participants (68 healthy controls, 58 diagnosed with schizophrenia). The proposed approach combines automatic speech transcription using the Truebar model with the extraction of verbal and non-verbal features, guided by established diagnostic criteria and enhanced by the generative capabilities of large language models. Feature relevance was assessed using a Bayesian statistical framework and subsequently used to train several classical machine learning models for schizophrenia classification. The primary contribution of this work is an Explainable Boosting Machine model that achieved a classification accuracy of 0.82 and an AUC of 0.90, effectively distinguishing between healthy controls and individuals with schizophrenia. In addition, the system provides both visual and textual explanations for each prediction, establishing it as the first fully automated and interpretable schizophrenia classification framework developed for the Slovene language.