Prediction of assistance dog training outcomes using machine learning and deep learning models

Abstract

This study investigates the predictive power of machine learning and deep learning models for forecasting training outcomes in assistance dogs, using behavioral survey data (C-BARQ) collected from volunteer puppy-raisers at two developmental stages: 6 months and 12 months. We used data from two assistance dog training organizations–Canine Companions and The Seeing Eye, Inc.– to assess model performance and generalizability across different training contexts. Six models, including traditional machine learning approaches (SVM, Random Forest, Decision Tree, and XGBoost) and deep learning architectures (MLP and CNN), were trained and evaluated on C-BARQ behavioral scores using metrics such as accuracy, F1 Score, precision, and recall. Results indicate that Support Vector Machine (SVM) and XGBoost consistently delivered the highest prediction accuracy, with SVM achieving up to 80 % accuracy in the Canine Companions dataset and 71 % in the Seeing Eye dataset. Although deep learning models like CNN showed moderate accuracy, traditional machine learning models excelled, particularly in structured, tabular data where feature separability is essential. Models trained on 12-month data generally yielded higher predictive accuracy than those trained on 6-month data, highlighting the value of extended behavioral observations. This research underscores the efficacy of traditional machine learning models for early-phase prediction and emphasizes the importance of aligning model selection with dataset characteristics and the stage of behavioral assessment.