| Diabetes is a chronic metabolic disease characterized by sustained high blood sugar levels above the normal range.Currently,data mining techniques can be used for its prediction,diagnosis,and treatment.Specific applications include high-risk patient screening,prediction of disease progression and complications risk,and optimization of treatment plans.These technologies help improve the effectiveness of diabetes management and become an important means of future diabetes management.However,most diabetes data mining techniques have not considered the issue of sharp points and the poor interpretability of mining results.This paper therefore examines and analyzes data mining techniques for diabetes,proposing fuzzy efficient pattern mining models and streaming fuzzy efficient pattern mining models.The main work of this paper is shown below:(1)A fuzzy efficient pattern mining model for the diabetes scenario is proposed.In order to solve the problem of poor interpretability of mining results in traditional efficient item set mining algorithm,fuzzy set theory is introduced and membership function is used.Then,the features in the data are blurred and the external utility values are set according to the importance of different features.In the mining process,both tree structure and list structure are used to calculate the overestimated utility value.Then,the fuzzy efficient use mining model combines the characteristics of the one-stage algorithm and the two-stage algorithm to effectively filter the fuzzy efficient use item set.Experiments show that the model can better reflect the meaning of the data itself,enhance the interpretability and understandability,and in some data scenarios,the model is better than most algorithms in terms of time and space efficiency.(2)A streaming fuzzy efficient pattern mining model suitable for the diabetes scenario is proposed.Aiming at the problem that the proposed fuzzy and efficient item set mining model can not be applied to the data flow scene,a model of fuzzy and efficient item set mining in data flow is designed and implemented by using the idea of sliding window.The model avoids processing the entire data stream by maintaining a dynamic sliding window,and uses filter trees to reduce the processing time of the candidate item set.In addition,the global Fuzzy Utility tree structure FHUIL-Tree(Fuzzy High Utility Items List-Tree)and FUDB(Fuzzy Utility Data Base)are proposed to store the data in the current window.As you swipe through the window,you can dynamically delete expired data and add new data in the tree and database.Experiments show that the model can mine highly interpretable fuzzy and efficient item sets in the data stream,and in some data scenarios,the spatio-temporal complexity of the model is lower than that of most algorithms.(3)Based on the fuzzy efficient pattern mining model and streaming fuzzy efficient pattern mining model,a prototype system for diabetes data mining was designed and implemented according to the actual needs of hospitals.The system adopts a B/S architecture with front-end and back-end separation,improving the stability and scalability of the system.It provides doctors with functions such as assisted diagnosis,data visualization,warning and alarm,etc.,to ensure that they can understand the health status of local diabetes patients in real-time and take timely response measures when problems occur. |
