Abstract
Respiratory infections are a major concern in pig farming as they negatively impact animal health and productivity. Coughing is a key symptom of respiratory disease and can be classified as productive or non-productive, but human assessment often leads to inconsistencies. This study aimed to use a machine learning model to classify pig coughs and investigate their association with respiratory infections. Cough sounds from 49 fattening pigs were recorded and analyzed using a Python-based machine learning system. The model's accuracy in detecting coughs was 0.72, compared to 0.69 for farmers. For classification of non-productive coughs, the machine learning results showed strong agreement with infection status by Mycoplasma hyopneumoniae, with a Spearman's correlation of 0.80 and a Cohen's Kappa of 0.79. However, the association with Porcine Circovirus type 2 was weak, with correlation and Kappa values of 0.05 and 0.037, respectively. These findings indicate that machine learning can classify pig coughs more accurately than human evaluators and that non-productive coughs are strongly linked to Mycoplasma infection but not to PCV2. This suggests the potential use of machine learning for more reliable disease monitoring and early detection in swine production.