Investigation of the possibilities of utilizing data mining methods in veterinary medicine through animal hospital electronic patient records

In veterinary medicine, the theoretical analysis of electronic patient records has received increasing attention in recent years, but the integration of these data into clinical practice and support for clinical decisions has been limited. This study was designed to test the applicability of data mining methods in different clinical scenarios using electronic patient records in order to transfer theoretical knowledge to practical use and to represent one of the efforts in this field. With the web-based applications developed within the scope of the study, it is aimed to provide steps to overcome the deficiency in practice. In line with this goal, the electronic patient records of 11173 patients admitted to the internal medicine clinic of Selçuk University Animal Hospital between 2021 and 2022 were selected as the material of the study. The methodology of the study was shaped using the CRISP-DM methodology, which has become a standard in recent years. Within the scope of the study, two different data mining applications were performed. In both applications, models were developed using random forest, XGBoost, support vector machines and artificial neural network algorithms. The performance of the models was compared with accuracy, precision, sensitivity, F1 score and AUC values and the best model was selected. In the first application, electronic patient records of 525 patients were used to classify gastroenteritis and parvoviral enteritis, which are common diseases in cats and dogs. The random forest model was selected as the best model with accuracy (0,87), precision (0,86), sensitivity (0,90), f1 score (0,88) and AUC (0,92). With the selected model, a web-based application was developed to classify gastroenteritis and parvoviral enteritis diseases. In the second application, models were developed with 1694 patient data labeled as infectious diseases, metabolic diseases and healthy. The model built with the XGBoost algorithm was selected as the model that best represents the data with 0,91 accuracy; 0,92 precision; 0,91 sensitivity, 0,91 F 1 score and 0,98 AUC value. A web-based application was developed to classify the disease type with the selected model. The findings of the study revealed that data mining methods show high performance over veterinary electronic patient records. These results show that data mining applications utilizing electronic patient records in veterinary medicine can be effectively applied in theoretical and practical fields. Key Words: CRISP-DM; Classification Algorithms; Veterinary Electronic Patient Records; Data Mining