Development of an AI-Driven Mobile Application for Poultry Disease Diagnosis and Decision Support Using a Pre-Trained Transformer-Based Model

Abstract

Poultry farming remains a major contributor to every nation’s economy, yet it is faced with many challenges, especially during disease outbreaks. In developing nations, the majority of poultry farmers lack formal knowledge in the field and access to extension agents. This article proposes the development of an AI-driven mobile application for poultry disease diagnosis and decision support that leverages a pre-trained transformer model. A dual architectural framework that integrates a Convolutional Neural Network-based image classifier and a transformer-based model was used for image-based disease detection and expert-level diagnosis and treatment recommendations, respectively. The application was developed using object-oriented analysis and design methodology. The results of the framework show that the best image-based classifier achieved 84.53% accuracy in classifying the listed poultry diseases. The transformer models also presented average precision accuracy (PA) of 95%, professional rating (PR) of 95%, a coherence rating (CR) of 97%, and a task success rate (TSR) of 96% based on the aggregation of independent assessments from 3 domain experts. These results prove that a mobile application can be engineered to democratize the possibilities associated with engaging machine learning (ML) in poultry farming since existing AI tools were not designed for illiterate farmers. Nevertheless, human veterinary professionals remain indispensable, especially in context-specific cases where clinical examination and consideration of birds’ health history are required to complement AI-driven insights.