| With the continuous development of artificial intelligence and other new technologies and the background of the era of big data,it is a very realistic problem to combine machine learning and other new technologies with traditional animal husbandry.The equine industry is a large part of traditional animal husbandry,so it is of great significance for the development of animal husbandry and the accurate medicine of horses to establish the prediction and evaluation system of horse health condition by using artificial intelligence combined with statistical means.In this paper,the medical data of horses were quantified,evaluated and predicted.The quantified characteristics were derived,and three groups of derived data could better reflect the future health status of horses,namely,Groupⅱ(disease manifestation at admission),Group ⅲ(physiological data at admission)and total score.The scores and correlation between the death group and the survival group were analyzed by receiver operating Characteristic(ROC)curve,and the scores were used as an important indicator for horse health evaluation.Using machine learning method to predict the health of the horse is for the sake of data from a scientific point of view affects the important characteristics of the horse health combined with the feature of horses on 18 applications of machine learning model to health for the future of the horses do a reasonable prediction of specific work can be summarized as the following:(1)Horse data completion.Data processing on the raw data will be deleted or fill the missing values and outliers,and for diseases not explicitly mentioned in the simplified acute physiological score,by querying data,consult prescriptions of the expert is quantified,such as grading,to gain the complete data sets for the subsequent analysis and model building of laid a foundation.(2)Feature engineering.Combined with previous studies,four independent variables were derived from the total data,and each variable was comprehensively considered according to the previously quantified data to explore the response degree of different scores in the death group and the survival group to the health status,so as to determine the optimal cut-off value.(3)Establishment of horse health prediction model.In the construction of the fusion model,we first use the relatively common machine learning classification model,including k-nearest neighbor model,logistic regression model,support vector machine model,decision tree model.Then the integration model is constructed,including random forest model,GBDT model and Light GBM model.All models are predicted on the original data set and the data after feature selection and parameter optimization.Four optimal machine learning models are selected to form the Stacking fusion model.The results showed that the best cut-off value of Group ⅲ was 15.5 points,the AUC value was0.842,the sensitivity was 83.6%,and the specificity was 74.6%,which could be used to evaluate the health degree of horses.The fusion model based on RF,GBDT,LGBM and logistic regression is superior to single model in original data,feature selection data,under-sampled balanced data set and over-sampled balanced data set.Model fusion and feature numerical quantification also provide reference for other prediction and evaluation studies. |
