Predicting guide dog career success using machine learning and large language models

Guide dogs play essential roles in supporting independence and well-being of visually impaired people. High attrition rates underscore the need for more effective early prediction of dog suitability for a guide dog career. This study presents a novel data-driven framework that integrates machine learning and large language models to predict guide dog training outcomes using data from the Israel Guide Dog Center (IGDC). The dataset included structured Behavior Checklist (BCL) assessments and unstructured trainer textual comments collected from 990 dogs at the pre-training stage. Incorporating trainer comments alongside structured behavioral assessments led to a substantial improvement in predictive performance compared to models based on BCL data alone. Furthermore, the pipeline generated interpretable textual explanations of model predictions, which were empirically evaluated by two experienced guide dog trainers. The findings highlight the potential of combining structured behavioral metrics with natural language data to enhance predictive accuracy and decision support in guide dog selection and training.