A MACHINE LEARNING FRAMEWORK FOR EARLY DETECTION AND DIAGNOSIS OF CANINE DIABETES

Abstract

Disease diagnosis in animals is often more challenging than in humans due to their inability to communicate their symptoms directly, and because many diseases exhibit similar clinical signs. Predicting the risk of diabetes in dogs is difficult because veterinary datasets are often noisy, imbalance, and contains heterogeneous clinical measurements. Machine learning based decision support systems offer an effective approach for analyzing such complex data to improve disease diagnostic accuracy. These systems can assist veterinary care providers in maintaining round-the-clock remote surveillance and enable veterinarians to have instant access to relevant patient information. This study presents a Canine Diabetes Diagnosis and Recommendation (CDDR) framework for predicting the severity of diabetes in canines using machine learning classifiers. Information Gain, a feature selection method, is used to identify the most relevant clinical and laboratory features, thereby reducing data dimensionality and improving model performance. Several machine learning algorithms, including Random Forest, LibSVM, Decision Stump, and REP Tree, were evaluated using 10-fold cross-validation. Among the evaluated classifiers within the CDDR framework, Random Forest achieves the highest accuracy of 93.0%, precision of 0.92, recall of 0.92, and the lowest mean absolute error of 0.07. Overall, our findings indicate that integrating feature selection, machine learning techniques, and decision support systems can significantly improve the accuracy and reliability of canine diabetes prediction. The proposed CDDR framework can assist veterinarians in early disease detection, clinical decision-making, and continuous remote monitoring of canine patients, especially in resource-constrained veterinary settings.

Keywords : 

Canines, Diabetes, Machine learning Algorithms, Classification, Feature ranking