Diabetic nephropathy classification model construction method and device based on fuzzy decision system

Abstract

The application discloses a diabetes nephropathy classification model construction method and device based on a fuzzy decision system. In the process of constructing the diabetes nephropathy classification model, the feature group generation strategy based on the fuzzy approximate Markov blanket (FAMB) is determined by introducing the Markov blanket algorithm and the approximation of the fuzzy rough set and the fuzzy uncertainty measure; the overlapping features are divided based on the attribution division mechanism of the fuzzy dependence degree; and finally the heterogeneous feature selection algorithm with group structure mining is determined to screen and optimize the feature groups. The application can focus on developing a high-efficiency calculation index, further utilizes the structural information among the features to improve the classification performance, and thus provides a reference for clinical auxiliary diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of neural networks, and in particular to a diabetic nephropathy classification model construction method and device based on a fuzzy decision system. BACKGROUND

[0002] In view of the great difference between diabetic nephropathy (DN) and non-diabetic renal disease (NDRD) in treatment and prognosis, the existing and disclosed solution technology is to diagnose and determine through "renal biopsy". However, due to its serious invasiveness and postoperative complications, this technology is limited in clinical practice. Therefore, it is of great significance to capture important pathological attributes using appropriate technology, mine biomarkers containing multiple attributes, and evaluate DN and establish a diagnostic prediction model through non-invasive diagnosis methods. SUMMARY

[0003] Based on this, the embodiments of the present application provide a diabetic nephropathy classification model construction method and device based on a fuzzy decision system, which can focus on developing a high computational efficiency measurement index, and further utilize the structural information between features to improve the classification performance, thereby providing a reference for clinical auxiliary diagnosis.

[0004] In a first aspect, a diabetic nephropathy classification model construction method based on a fuzzy decision system is provided, which includes:

[0005] According to the target diabetic nephropathy data set, the diabetic nephropathy detection indexes are determined; the data corresponding to each diabetic nephropathy detection index is imported into the SPSS software, and the data is preprocessed; and the missing values in the data set are processed using the average value method of adjacent points; according to the preprocessed biopsy data, the data is divided into a DN group data set and an NDRD group data set;

[0006] A diabetic nephropathy classification model is constructed; wherein the diabetic nephropathy classification model specifically includes the introduction of Markov blanket algorithm and fuzzy rough set approximation and fuzzy uncertainty measure, determination of feature group generation strategy based on FAMB; further division of overlapping features based on fuzzy dependency attribution mechanism; and finally determination of heterogeneous feature selection algorithm with group structure mining, screening and optimization of feature groups;

[0007] The diabetic nephropathy classification model is trained through the divided DN group data set and NDRD group data set until a trained diabetic nephropathy classification model is obtained.

[0008] Optionally, the method further includes:

[0009] The constructed diabetic nephropathy classification model is compared with other feature selection algorithm models; the other feature selection algorithm models at least include DRGS, FastCFS, FCBF, IWFS, MRI, MRMI and N-MRMCR-MI algorithms;

[0010] and a classification result is determined through multiple classifiers; the classifiers at least include Classification AndRegression Tree, Bayes, K-Nearest Neighbor and Support Vector Machine classifiers;

[0011] Based on the classification accuracy and the average reduction rate, evaluation results of each model are obtained.

[0012] Optionally, the model comparison and evaluation are realized through Matlab software, and a relationship diagram of the selected feature quantity and the classification accuracy obtained by different algorithm models is generated.

[0013] Optionally, before the diabetic nephropathy classification model is trained through the divided DN group data set and NDRD group data set, the method further comprises:

[0014] The data set is divided into training / test data sets through cross-validation.

[0015] Optionally, according to the target diabetic nephropathy data set, the diabetic nephropathy detection indexes are determined to include:

[0016] ID, physical indexes, diagnosis results, prealbumin, creatine kinase, urea, apolipoprotein A1, low-density lipoprotein cholesterol, albumin / globulin, glomerular filtration rate, alkaline phosphatase, apolipoprotein B, amylase, alpha-hydroxybutyric acid dehydrogenase, V-glutamyl transpeptidase, indirect bilirubin, total calcium, potassium and chlorine of the patient.

[0017] In a second aspect, a diabetic nephropathy classification model construction device based on a fuzzy decision system is provided, and the device comprises:

[0018] A preprocessing module is configured to analyze and determine diabetic nephropathy detection indexes according to a target diabetic nephropathy data set; import data corresponding to each diabetic nephropathy detection index into SPSS software, and preprocess the data; process missing values in the data set by using a nearest point average method; divide the data into a DN group data set and an NDRD group data set according to the preprocessed biopsy data;

[0019] The constructing module is configured to construct a diabetic nephropathy classification model; wherein the diabetic nephropathy classification model specifically comprises an introduction of Markov blanket algorithm and fuzzy rough set approximation and fuzzy uncertainty measure, determination of a feature group generation strategy based on FAMB, further division of overlapping features based on a fuzzy dependence degree attribution division mechanism, and finally determination of a heterogeneous feature selection algorithm with group structure mining to screen and optimize the feature groups;

[0020] The training module is configured to train the diabetic nephropathy classification model through the divided DN group data set and NDRD group data set until a trained diabetic nephropathy classification model is obtained.

[0021] Optionally, the device further comprises:

[0022] The comparison module is configured to compare the constructed diabetic nephropathy classification model with other feature selection algorithm models; wherein the other feature selection algorithm models at least include DRGS, FastCFS, FCBF, IWFS, MRI, MRMI and N-MRMCR-MI algorithms; and a classification result is determined through multiple classifiers; wherein the classifiers at least include ClassificationAndRegressionTree, Bayes, K-Nearest Neighbor and SupportVectorMachine classifiers; and evaluation results of the models are obtained based on classification accuracy and average reduction rate.

[0023] Optionally, the model comparison and evaluation are realized through Matlab software, and a relationship diagram of the number of selected features and classification accuracy of different algorithm models is generated.

[0024] Optionally, before training the diabetic nephropathy classification model through the divided DN group data set and NDRD group data set, the device further comprises:

[0025] The data set is divided into training / test data sets through cross-validation.

[0026] Optionally, according to the target diabetic nephropathy data set, the diabetic nephropathy detection indexes are determined to specifically include:

[0027] ID, physical indexes and diagnosis results of the patient, prealbumin, creatine kinase, urea, apolipoprotein A1, low-density lipoprotein cholesterol, albumin / globulin, glomerular filtration rate, alkaline phosphatase, apolipoprotein B, amylase, alpha-hydroxybutyric acid dehydrogenase, V-glutamyl transpeptidase, indirect bilirubin, total calcium, potassium and chlorine.

[0028] The technical scheme provided by the embodiment of the application constructs a diabetic nephropathy classification model, specifically determines a feature group generation strategy based on FAMB by introducing a Markov blanket algorithm and approximation and fuzzy uncertainty measurement of a fuzzy rough set, further divides the overlapping features based on a fuzzy dependence attribution division mechanism, and finally determines a heterogeneous feature selection algorithm with group structure mining to screen and optimize the feature groups. It can be seen that the beneficial effects of the application are that a high-computing-efficiency measurement index can be developed, and the structural information between features is further utilized to improve the classification performance, thereby providing a reference for clinical auxiliary diagnosis. BRIEF DESCRIPTION OF DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the application or the technical solutions in the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only exemplary, and those skilled in the art can also obtain other implementation drawings according to the provided drawings without creative labor.

[0030] Figure 1 A diabetic nephropathy classification model construction flowchart based on a fuzzy decision system is provided for the embodiment of the application.

[0031] Figure 2 A flowchart of the HFS-GSM algorithm is provided for the embodiment of the application.

[0032] Figure 3 A HFS-GSM algorithm framework diagram is provided for the embodiment of the application. DETAILED DESCRIPTION

[0033] In order to make the purposes, technical solutions and advantages of the application more clear, the application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the application and not to limit the application.

[0034] In the description of the application, the terms "comprising", "having" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to only those steps or units explicitly listed, but can also include other steps or units inherent to the process, method, product or device, or steps or units added based on the inventive concept further optimization scheme.

[0035] In view of the great difference between diabetic nephropathy (DN) and non-diabetic renal disease (NDRD) in treatment and prognosis, the existing and disclosed solution is to diagnose by "renal biopsy". However, due to its serious invasiveness and postoperative complications, the technology is limited in clinical practice. Therefore, it is of great significance to evaluate DN and establish a diagnostic prediction model by non-invasive diagnosis method by using appropriate technology to capture important pathological attributes and mine biomarkers containing multiple attributes.

[0036] Defects in the extraction of existing diabetic nephropathy detection indicators: By using different classifiers in 12 public data sets, the extraction of diabetic nephropathy detection indicators is carried out with other 7 representative feature selection algorithm models. It is found that the defects of the existing solution are that MRMI obtains the best reduction result when tested with four different classifiers. However, this is at the expense of a substantial sacrifice of classification accuracy. Similarly, the FCBF algorithm with output subset does not obtain a good trade-off in terms of classification accuracy and the number of selected features, although its classification accuracy is slightly better than that of the MRMI algorithm in some cases. The remaining other algorithms have the problem of large original space redundancy.

[0037] Specifically, refer to Figure 1 which shows a flowchart of a diabetic nephropathy classification model construction method based on a fuzzy decision system provided by the embodiments of the present application. The method can include the following steps:

[0038] Step 101, according to the target diabetic nephropathy data set, the diabetic nephropathy detection indicators are determined by analysis.

[0039] The data corresponding to each diabetic nephropathy detection indicator is imported into the SPSS software, and the data is preprocessed; and the missing values in the data set are processed by using the nearest point average value method; according to the preprocessed biopsy data, the data is divided into a DN group data set and an NDRD group data set.

[0040] Step 102, constructing a diabetic nephropathy classification model.

[0041] The diabetic nephropathy classification model specifically includes introducing Markov blanket algorithm and fuzzy rough set approximation and fuzzy uncertainty measure, determining feature group generation strategy based on FAMB; further dividing the overlapping features based on fuzzy dependence degree attribution division mechanism; and finally determining the heterogeneous feature selection algorithm with group structure mining to screen and optimize the feature groups.

[0042] Step 103, training the diabetic nephropathy classification model through the divided DN group data set and NDRD group data set until the training of the diabetic nephropathy classification model is completed.

[0043] The HFS-GSM framework is divided into three parts: first, the running mechanism of the feature group generator is based on the concept of fuzzy approximation Markov blanket with fuzzy uncertainty measure, which plans the feature group generation strategy for group structure mining; second, for the case of feature group overlap, an overlapping group elimination mechanism is developed, in which fuzzy dependency is used to divide the features in the overlapping area to determine which group they belong to; finally, in the feature group selection stage, a heterogeneous feature selection algorithm with group structure mining is designed.

[0044] The HFS-GSM framework is divided into three parts: first, the running mechanism of the feature group generator is based on the concept of fuzzy approximation Markov blanket with fuzzy uncertainty measure, which plans the feature group generation strategy for group structure mining; second, for the case of feature group overlap, an overlapping group elimination mechanism is developed, in which fuzzy dependency is used to divide the features in the overlapping area to determine which group they belong to; finally, in the feature group selection stage, a heterogeneous feature selection algorithm with group structure mining is designed.

[0045] Using the idea of approximation and Markov blanket, this method not only utilizes group structure mining information, but also improves computational efficiency to some extent. This is because it replaces the global computation based on probability calculation with the return. In addition, a mechanism for eliminating overlapping groups is established using fuzzy dependency, which avoids the redundancy that may exist in the calculation process. By comparing with 7 representative feature selection methods in 12 public medical data sets, the superiority and effectiveness of the method are proved. More importantly, through the examination of a real medical case of diabetic nephropathy, the effectiveness and feasibility of the method in medical diagnosis are further illustrated from practical application. It provides a good solution for doctors to make clinical auxiliary diagnosis.

[0046] The method in the application also includes the comparison process with other algorithms, and a specific embodiment is given as follows:

[0047] (1) Experimental data preprocessing

[0048] ① Preliminary preparation

[0049] Dataset: According to the hospital-provided dataset of diabetic nephropathy, the detection indicators of diabetic nephropathy were analyzed, such as the ID of the patient, physical indicators, diagnosis results, prealbumin, creatine kinase, urea, urea, apolipoprotein A1, low-density lipoprotein cholesterol, low-density lipoprotein cholesterol, albumin / globulin, glomerular filtration rate (eGFR), alkaline phosphatase (AKP), apolipoprotein B (Apo), amylase (AMY), alpha-hydroxybutyric acid dehydrogenase, V-glutamyl transpeptidase, indirect bilirubin, total calcium, potassium, chlorine, etc.

[0050] ②Experimental process

[0051] The dataset was imported into SPSS software; first, the repeated and missing data were preprocessed; then the "average value of adjacent points" method was used to process the missing values in the dataset.

[0052] ③Experimental results

[0053] A total of 1367 complete case entries were obtained, including 54 test indicators. According to the pathological biopsy results, the patients were divided into the DN group (127 cases) and the NDRD group (1240 cases).

[0054] (2) Screening of feature indicators

[0055] ①Experimental data and related specifications

[0056] Dataset: effective dataset after experimental preprocessing;

[0057] Algorithm model: DRGS, FastCFS, FCBF, IWFS, MRI, MRMI and N-MRMCR-MI, HFS-GSM, etc. feature selection algorithm model;

[0058] Evaluation specification: classification accuracy based on four classifiers Classification And Regression Tree (CART), Bayes (NB), K-Nearest Neighbor (KNN), Support Vector Machine (SVM), when tested on four different classifiers, the classification accuracy and average reduction rate produced by the eight algorithms on the DN dataset were set as evaluation indicators. Experimental specification: the higher the classification accuracy and average reduction rate, the better the algorithm effect;

[0059] ②Experimental process

[0060] Selection of feature indicators: the dataset was put into the feature selection algorithm respectively, and the feature selection was carried out;

[0061] ​Algorithm effect evaluation: according to the agreed algorithm effect evaluation specification, each feature selection algorithm is evaluated respectively;

[0062] Select the optimal algorithm: select the optimal algorithm "HFS-GSM algorithm" according to the algorithm effect evaluation, and the classification accuracy is "91.95+0.24", and the average reduction rate is "74.0175%";

[0063] ③ Experimental results

[0064] The experimental results show that the following indicators "classification accuracy, average reduction rate" are selected as characteristic indicators.

[0065] The technical principle of this embodiment is specifically: the concept of Markov blanket is introduced in fuzzy rough set, and a feature group generation strategy based on fuzzy approximate Markov blanket (FAMB) is designed by using the ideas of approximation and fuzzy uncertainty measure. In addition, an attribution division mechanism based on fuzzy dependency is proposed to solve the problem of feature group overlap.

[0066] Through the use of different classifiers in 12 public medical data sets, the extraction of diabetic nephropathy detection indicators is carried out with other 7 representative feature selection algorithm models, and then the algorithm model for predicting early diabetic nephropathy patients is constructed and the prediction performance of each algorithm model is compared. The best model is selected to assist clinical decision-making.

[0067] In the implementation process of this study, the data dimension reduction performance of DRGS, FastCFS, FCBF, IWFS, MRI, MRMI, N-MRMCR-MI and HFS-GSM algorithm models is proved by a large number of experiments, and the accuracy of the diabetic nephropathy feature indicators extracted by several algorithm models is evaluated. Experimental data is divided into 12 public medical data sets (audiology, cardiotolography, cleveland, covid-19, dermatology, HCV, hepatitis, lung cancer, lymphography, parkinsons, SPECT, thyroid disease) and diabetic nephropathy patient classification (DN) data sets. These data are widely used in practical applications, and meet the conditions of different sources, different categories, different features and different sample sizes. The data set includes discrete features and continuous features, and the fuzzy rough set tool is used to reduce the data dimension of discrete and continuous features at the same time in the experiment.

[0068] Experimental Setup: This experiment compares eight feature selection algorithms: DRGs, FastCFS, FCBF, IWFS, MRI, MRMI, N-MRMCR-MI, and HFS-GSM. Parameters were set according to the different characteristics of each algorithm and the experimental requirements. For the classifier settings, four different classifiers were selected (Classification And Regression Tree (CART)). The experiments employed three algorithms: Bayes (NB), K-Nearest Neighbor (KNN), and SupportVectorMachine (SVM). First, the dataset was divided into training and testing datasets using cross-validation. Second, the feature subsets were determined and configured. Next, the classifier was trained on the training dataset, and then the test data was input into the classifier to obtain the classification accuracy. Finally, the relationship between the number of selected features and the classification accuracy was plotted using different algorithm models. The experiments were implemented using Matlab software.

[0069] The experimental results were evaluated using classification accuracy and average reduction rate. Four different classifiers were used to assess classification accuracy. The matching rate of the feature selection results was indirectly represented by professional metrics provided by medical experts and the average reduction rate extracted based on the algorithm model.

[0070] Feature extraction was performed on both the benchmark dataset and the diabetic nephropathy patient classification dataset. The experimental results are as follows:

[0071] The classification accuracy tables for each algorithm based on four classifiers (CART, NB, KNN, and SVM) are shown in Tables 1, 2, 3, and 4. Since the feature selection algorithm sorts the weights of each indicator (Table 5), the average reduction rate (%) of the eight algorithms tested on all datasets and the DN dataset using four different classifiers was calculated. The average reduction rates of the HFS-GSM algorithm were 70%, 69%, 65%, 63% and 73.1%, 79.52%, 73.1%, 70.35%, respectively. This shows that although the feature classification accuracy extracted by the HFS-GSM algorithm is not as good as that of the MRMI algorithm, it significantly reduces the consumption of computational resources and ensures overall classification performance. Specifically, the feature indicators are: classification accuracy "91.95 + 0.24", and average reduction rate "74.0175%".

[0072] Table 1. Comparison of classification performance of different algorithms using CART.

[0073]

[0074] Table 2 The classification performance of different algorithms with NB

[0075]

[0076] Table 3 The classification performance of different algorithms with KNN

[0077]

[0078] Table 4 The classification performance of different algorithms with SVM

[0079]

[0080] Table 5 The classification performance of different algorithms on DN dataset

[0081]

[0082] The embodiment of the application further provides a diabetic nephropathy classification model construction device based on a fuzzy decision system, comprising:

[0083] A preprocessing module is configured to analyze and determine diabetic nephropathy detection indexes according to a target diabetic nephropathy data set, import data corresponding to each diabetic nephropathy detection index into SPSS software, and preprocess the data; missing values in the data set are processed by using a nearest point average method; and according to the biopsy data after preprocessing, the data is divided into a DN group data set and a NDRD group data set.

[0084] A construction module is configured to construct a diabetic nephropathy classification model; wherein the diabetic nephropathy classification model specifically comprises an introduction of a Markov blanket algorithm and fuzzy rough set approximation and fuzzy uncertainty measure, determination of a feature group generation strategy based on FAMB, further division of overlapping features based on a fuzzy dependence attribution division mechanism, and finally determination of a heterogeneous feature selection algorithm with group structure mining, screening and optimization of the feature group.

[0085] a training module configured to train the diabetic nephropathy classification model by using the divided DN group dataset and NDRD group dataset until a trained diabetic nephropathy classification model is obtained.

[0086] In an optional embodiment of the present application, the device further comprises:

[0087] a comparison module configured to compare the constructed diabetic nephropathy classification model with other feature selection algorithm models, wherein the other feature selection algorithm models at least include DRGS, FastCFS, FCBF, IWFS, MRI, MRMI and N-MRMCR-MI algorithms; and determine a classification result by using a plurality of classifiers, wherein the classifiers at least include ClassificationAndRegressionTree, Bayes, K-Nearest Neighbor and SupportVectorMachine classifiers; and obtain evaluation results of the models based on classification accuracy and average reduction rate.

[0088] In an optional embodiment of the present application, the model comparison and evaluation are realized by using Matlab software, and a relationship diagram between the number of selected features and classification accuracy of different algorithm models is generated.

[0089] In an optional embodiment of the present application, before the diabetic nephropathy classification model is trained by using the divided DN group dataset and NDRD group dataset, the method further comprises:

[0090] dividing the dataset into training / test datasets by using cross-validation.

[0091] In an optional embodiment of the present application, according to the target diabetic nephropathy dataset, the diabetic nephropathy detection indexes are determined to include at least:

[0092] ID, physical indexes and diagnosis results of the patient, prealbumin, creatine kinase, urea, apolipoprotein A1, low-density lipoprotein cholesterol, albumin / globulin, glomerular filtration rate, alkaline phosphatase, apolipoprotein B, amylase, alpha-hydroxybutyric acid dehydrogenase, V-glutamyl transpeptidase, indirect bilirubin, total calcium, potassium and chlorine.

[0093] The device for constructing a diabetic nephropathy classification model based on a fuzzy decision system is used for implementing the method for constructing a diabetic nephropathy classification model based on a fuzzy decision system. Specific limitations of the device for constructing a diabetic nephropathy classification model based on a fuzzy decision system can be found in the limitations of the method for constructing a diabetic nephropathy classification model based on a fuzzy decision system, which will not be described here again. Each part of the device for constructing a diabetic nephropathy classification model based on a fuzzy decision system can be realized by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a hardware form, or can be stored in a memory in a software form in a device, so as to be called and executed by a processor to perform operations corresponding to each module.

[0094] Each technical feature of the above-described embodiments can be combined arbitrarily. To make the description simple, all possible combinations of each technical feature in the above-described embodiments are not described, but as long as the combinations of the technical features do not exist, they should be considered as the scope of the present disclosure.

[0095] The above-described embodiments only express several implementation manners of the present disclosure, and the description is relatively specific and detailed, but it should not be understood as a limitation on the patent scope of the present disclosure. It should be pointed out that, for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present disclosure, and these all belong to the protection scope of the present disclosure. Therefore, the patent protection scope of the present disclosure should be subject to the appended claims.

Claims

1. A method for constructing a classification model for diabetic nephropathy based on a fuzzy decision system, characterized in that, The method includes: Based on the target diabetic nephropathy dataset, the detection indicators for diabetic nephropathy were determined through analysis. The data corresponding to each diabetic nephropathy detection indicator were imported into SPSS software and preprocessed. Missing values ​​in the dataset were processed using the nearest neighbor mean method. Based on the preprocessed biopsy data, the data were divided into DN group dataset and NDRD group dataset. A classification model for diabetic nephropathy is constructed. Specifically, this model incorporates the Markovblanket algorithm, fuzzy rough set approximation, and fuzzy uncertainty measure to determine a feature group generation strategy based on fuzzy approximation Markovblanket. Furthermore, an overlapping group elimination mechanism is established based on fuzzy dependency, using fuzzy dependency relationships to partition features in overlapping regions to determine the group to which a feature belongs. Finally, a heterogeneous feature selection algorithm with group structure mining is determined to filter and optimize the feature groups. The diabetic nephropathy classification model was trained using the divided DN group dataset and NDRD group dataset until a fully trained diabetic nephropathy classification model was obtained. The method also includes: The constructed diabetic nephropathy classification model was compared with other feature selection algorithm models; among which, other feature selection algorithm models included at least DRGs, FastCFS, FCBF, IWFS, MRI, MRMI and N-MRMCR-MI algorithms. The classification result is determined by using multiple classifiers; among which, the classifiers include at least the Classification And Regression Tree, Naïve Bayes, K-Nearest Neighbor, and Support Vector Machine classifiers. The evaluation results of each model are obtained based on classification accuracy and average reduction rate; The model comparison and evaluation were carried out using Matlab software, and the relationship between the number of selected features and the classification accuracy was obtained by generating different algorithm models. Before training the diabetic nephropathy classification model using the divided DN group dataset and NDRD group dataset, the method further includes: The dataset is split into training and testing datasets using cross-validation.

2. The method for constructing a classification model for diabetic nephropathy according to claim 1, characterized in that, Based on the target diabetic nephropathy dataset, the specific indicators for detecting diabetic nephropathy were determined through analysis: Patient ID, physical indicators, diagnosis results, prealbumin, creatine kinase, urea, apolipoprotein A1, low-density lipoprotein cholesterol, albumin / globulin ratio, glomerular filtration rate, alkaline phosphatase, apolipoprotein B, amylase, α-hydroxybutyrate dehydrogenase, V-glutamyl transpeptidase, indirect bilirubin, total calcium, potassium, and chloride.

3. A device for constructing a diabetic nephropathy classification model based on a fuzzy decision system, characterized in that, The device includes: The preprocessing module is used to analyze and determine the detection indicators of diabetic nephropathy based on the target diabetic nephropathy dataset; import the data corresponding to each diabetic nephropathy detection indicator into SPSS software and preprocess the data; process missing values ​​in the dataset using the nearest neighbor mean method; and divide the data into DN group dataset and NDRD group dataset based on the preprocessed biopsy data. A construction module is used to build a classification model for diabetic nephropathy. Specifically, the classification model for diabetic nephropathy includes introducing the Markov blanket algorithm and approximation with fuzzy rough sets and fuzzy uncertainty measures to determine a feature group generation strategy based on fuzzy approximation Markov blanket. Furthermore, an overlapping group elimination mechanism is established based on fuzzy dependencies, and fuzzy dependencies are used to divide features in overlapping regions to determine the group to which the features belong. Finally, a heterogeneous feature selection algorithm with group structure mining is determined to screen and optimize the feature groups. The training module is used to train the diabetic nephropathy classification model using the divided DN group dataset and NDRD group dataset until a fully trained diabetic nephropathy classification model is obtained. The device also includes: The comparison module is used to compare the constructed diabetic nephropathy classification model with other feature selection algorithm models. These other feature selection algorithm models include at least DRGs, FastCFS, FCBF, IWFS, MRI, MRMI, and N-MRMCR-MI algorithms. The module also determines the classification results using multiple classifiers, including at least Classification And Regression Tree, Naïve Bayes, K-Nearest Neighbor, and Support Vector Machine classifiers. Evaluation results for each model are obtained based on classification accuracy and average reduction rate. The model comparison and evaluation were carried out using Matlab software, and the relationship between the number of selected features and the classification accuracy was obtained by generating different algorithm models. Before training the diabetic nephropathy classification model using the divided DN group dataset and NDRD group dataset, the following steps are also included: The dataset is split into training and testing datasets using cross-validation.

4. The device for constructing a classification model for diabetic nephropathy according to claim 3, characterized in that, Based on the target diabetic nephropathy dataset, the specific indicators for detecting diabetic nephropathy were determined through analysis: Patient ID, physical indicators, diagnosis results, prealbumin, creatine kinase, urea, apolipoprotein A1, low-density lipoprotein cholesterol, albumin / globulin ratio, glomerular filtration rate, alkaline phosphatase, apolipoprotein B, amylase, α-hydroxybutyrate dehydrogenase, V-glutamyl transpeptidase, indirect bilirubin, total calcium, potassium, and chloride.

Images