Prediction Of Urinary Tract Infections Based On Improved Gradient Boosting Decision Tree

Urinary tract infection is a common disease in the aged-population,whose risk increases with age.Particularly,due to physical conditions,the elderly or disabled people are more likely to develop urinary tract infections than other groups.At present,the detection methods for urinary tract infections in hospitals,especially for the elderly or people with limited mobility,are mainly through routine urine tests or catheterization.The routine urine test has to be conducted in hospitals or clinical centers,which is of inconvenience for the elderly and disabled groups.Besides,urinary catheters are likely to increase the risk of urinary tract infections.The laboratory where the author is affiliated has developed a set of intelligent point-care-of urinary test system that can be embedded in diaper,which can effectively solve the challenges of sample collection and immediate urinary test for the forgoing groups.The work of this thesis is part of the research of this intelligent urinary test system.Based on this system,the modeling and experimental verification are carried out.The intelligent urine test system includes colorimetric detection of five urine biological indicators including blood,nitrite,glucose,leukocyte and protein.After using the urine urinary test equipment to perform sampling and real-time inspection,the smartphone camera is used to collect the colorimetric result pictures for simple preprocessing.The RGB channel values of the result pictures as the feature vectors of biological indicators are read through MATLAB programming.The K-nearest neighbor model was used to classify these five biological indicators,which are compared with control models such as support vector machine,random forest,and classification and regression tree.The results show that the K-nearest neighbor model is more suitable for the data set.After analyzing and obtaining the classification results of the five biological indicators,a new data set characterized by the result class markers of the five biological indicators is formed.Due to the imbalance of samples in the two categories of illness and health in this data set,this thesis develops the cost-sensitive learning method to improve the gradient lifting decision tree algorithm,and applies this model to predict whether the patient suffers from urinary tract infection disease.In this work,the gradient lifting decision tree,random forest,support vector machine and naive Bayes model are used to comprehensively analyze the dimensions of the modeling duration,confidence interval,confusion matrix and ROC curve area and the improved gradient lifting decision tree model.The result proves that the improved gradient lifting decision tree model is optimal,and the prediction accuracy of urinary tract infection diseases reaches 94.7%,which is about 3% higher than the original gradient lifting decision tree model,showing its better classification capability.The model can be applied to the auxiliary detection software of the intelligent urinary test system,which is helpful to realize the rapid pre-diagnosis of diseases related to urinary tract infection.