DEVELOPMENT OF MACHINE LEARNING MODEL FOR A DOG DISEASES EXPERT SYSTEM

Abstract

Junior veterinarians often make the mistake of misdiagnosing animals during their training years. Senior veterinarians are usually too busy to assist them every time. To solve that problem, an expert system which implements machine learning model that can be used to identify dog diseases based on given symptoms is proposed to help junior veterinarians. This expert system will be in the form of web-application which is built using ReactJS. The machine learning model is trained using five algorithms (Logistic Regression (LR), Naïve Bayes (NB), K-Nearest Neighbor (KNN), and Support Vector Machine (SVM)) using multi-label classification task and ROC-AUC score is used as an evaluation metric. Hyperparameter tuning was also performed to improve the performance for each model. The dataset used for the training has five different labels. After conducting the training, the best algorithm for each label are: NB for label A (0.78), SVM for label B (0.68), KNN for label C (0.70), LR for label D (0.96), and SVM for label E (0.82), with Random Forest having the longest training time and Naïve Bayes as the shortest. These models are then exported and hosted using a flask server so it can be accessed from the website by calling the API.