Research On Diabetes Decision-making Algorithm Based On Deep Learning

Diabetes has a large number of patients worldwide,it is extremely harmful to humans,and the mortality rate has risen sharply in recent years.At the same time,diabetes can cause a variety of serious complications,such as high blood pressure,gastrointestinal diseases,diabetic eye disease,vascular disease and so on.There is no effective treatment for diabetes currently.Therefore,early detection of diabetes is the key to reducing mortality among diabetic patients.In recent years,with deep learning yielding unusually brilliant results in the fields of image recognition,natural language understanding,automatic question and answer,machine translation,sentiment analysis,and stock price forecasting,researchers have begun to apply deep learning to medical diagnostics.Diabetes is one of the most common non-communicable diseases,and the study of deep learning applied to diabetes prediction is more worthy of attention.At present,scholars generally use traditional machine methods when studying the prediction of diabetes.In view of the current research status of diabetes prediction and the development trend of deep learning,this paper will study the diabetes prediction technology based on the deep learning platform TensorFlow.The main work has the following four aspects:1)The theory of data preprocessing is deeply studied,and the data preprocessing of Pima Indian diabetes data set from UCI is carried out.Emphasis is laid on the application of using average value to replace the missing value and feature selection using information gain algorithm(IG)to complete the preprocessing of diabetes data set.2)The effects of K-Nearest Neighbor(KNN)and Gradient Boosting Decision Tree(GBDT)on diabetes prediction are studied,and we analyze those two common classification algorithms' performance evaluation and prediction results.3)Study the specific implementation of deep neural network(DNN)in open source artificial deep learning system.TensorFlow,a deep learning platform based on Google's open source,builds a deep neural network(DNN)as a classifier based on forward propagation algorithm and BP algorithm,and then tests the effects between the deep neural algorithm and traditional machine learning method.4)Based on the above research results,we build a diabetes prediction and diagnosis platform.Starting from the actual needs of the hospital,we use java language to realize a prediction platform including data preprocessing and deep neural network as a classifier,providing a user-friendly and interactive platform for doctors and trying to provide reliable diagnostic results for the doctors.The results we get from experiments illustrate that the prediction model built on the deep neural network(DNN)algorithm has certain advantages compared with the traditional machine learning prediction model.As the data set increases,the advantages of the deep neural network model will become more apparent,and the accuracy of prediction can be continuously improved.At present,the solution to build a deep neural network using TensorFlow is relatively mature.