Abstract

Although canine and humans have interacted for centuries, human understanding of canine health remains limited. Several canine -health-monitoring systems have been developed in recent times. However, several factors, such as the canine breed, age, size, and weight, make data estimation challenging. In this study, activity sensors were installed on the necks of 30 canines of six breeds, ages, and weights, and a novel disease-inference program was developed to track changes in their scratching, licking, swallowing, and sleeping behaviors. Further, health questionnaires were created for similar diseases based on the observed abnormal canine behavior. In addition, a software program was developed and verified to predict canine diseases based on these data and recommend check-ups accordingly. The sensitivity and specificity of decision-making were verified by comparing the data on behavioral pattern changes and disease predictions collected via questionnaires with the results of veterinarian diagnoses. The average sensitivity and specificity of disease predictions (digestive and skin), estimated by the changes in behavioral patterns and the owner questionnaire, were 82% and 81%, respectively. Cohen's kappa coefficient was 0.79 in the diagnostic area, demonstrating diagnosis consistency. Therefore, the results show that canine’s' abnormal scratching, licking, swallowing, and sleeping patterns can be used for health monitoring. This study contributes to the development of canine health status monitoring systems.