Diabetes risk assessment model construction method based on deep learning
Through the deep learning-based diabetes risk assessment model, important feature parameters are selected, noise is removed and the characteristic parameter similarity matrix is formed, which solves the misdiagnosis and misdiagnosis of diabetes diagnosis in the prior art, and improves the accuracy and predictive ability of diabetes risk assessment.
Patent Information
- Application Number
- CN202411572697.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-06
- Publication Date
- 2025-01-10
AI Technical Summary
The existing diabetes diagnosis methods rely on doctor experience, have misdiagnosis and misdiagnosis, and the generalization ability of a single global optimization model is weak and has poor fault tolerance, making it difficult to effectively predict diabetes risk.
Using a diabetes risk assessment model based on deep learning, important feature parameters are selected through the data screening unit, and the data processing unit normalizes and noise removal, and a deep learning unit is constructed to form a characteristic parameter similarity matrix to predict diabetes risk.
It improves the accuracy and prediction ability of the diabetes risk assessment model, reduces the complexity of the model, avoids overfitting, enhances the resistance to noise, and improves the scientific nature of the diagnosis.
Smart Images
Figure CN119296805A_ABST
Abstract
Description
Technical Field
[0001] The present invention eliminates a method for constructing a diabetes risk assessment model based on deep learning, and relates to the technical field of model construction. Background Art
[0002] In recent years, with the rapid development of my country's economy and the changes in the national lifestyle and dietary structure, the number of diabetes patients in my country is growing at an alarming rate. At present, most medical institutions still rely on the doctor's personal experience and physical examination data for the diagnosis of diabetes. This diagnostic method has high requirements for doctors and is highly subjective. It is easy to misdiagnose and miss the best time for prevention and treatment, which will seriously affect the patient's physical and mental health. If diabetes is combined with machine learning and machine learning algorithms are used to assist doctors in diagnosis, it will greatly improve the diagnosis and scientificity, and effectively overcome the subjective problem of doctors' diagnosis based on experience. Therefore, it is of great practical significance to use machine learning methods to predict diabetic patients, such as logistic regression, BP neural network model, COX proportional risk model, and decision tree model. These methods have good results in disease prediction, but they also have various disadvantages, such as requiring a large amount of data, requiring continuous data for a certain period of time, weak generalization ability, overfitting, falling into local minimum, insensitivity to random and volatile data, and unsatisfactory prediction effects on unbalanced data. Most existing studies use a single global optimization model. The single classifier model has limited performance and has problems such as weak generalization ability and poor fault tolerance. Summary of the invention
[0003] In order to solve the above technical problems, the present invention proposes a method for constructing a diabetes risk assessment model based on deep learning, comprising the following steps: S1. Constructing a data screening unit, wherein the data screening unit performs feature selection by calculating contribution values, and screens out a plurality of important feature parameters that have evaluation value for diabetes risk; S2. Constructing a data processing unit, wherein the data screening unit obtains monitoring data of each important characteristic parameter at multiple monitoring time points, normalizes the monitoring data, and obtains a normalized data sequence of each characteristic parameter; S3, constructing a parameter forming unit, wherein the parameter forming unit calculates the noise index of the normalized data sequence of each characteristic parameter, and assigns weights to the plurality of characteristic parameters according to the noise index; S4. Construct a deep learning unit, wherein the deep learning unit removes noise data from the normalized data sequence of each characteristic parameter to obtain a normalized data sequence after the noise data is removed, performs deep learning based on the normalized data sequence after the noise data is removed, and outputs a diabetes risk type.
[0004] Furthermore, in step S2, the monitoring data of each important characteristic parameter I at multiple monitoring time points are normalized so that the monitoring data are mapped to the interval [-1, 1] to obtain normalized data : , t=1,2,…T; Among them, x It is the monitoring data of the t-th monitoring time point of characteristic parameter I, T is the total number of monitoring time points, x max is the maximum data value in the monitoring data, x min is the minimum data value in the monitoring data, Represents the normalized data sequence of feature parameter I { }.
[0005] Furthermore, the step S3 comprises: S31, constructing a noise identification submodule, wherein the noise identification submodule identifies noise data of a normalized data sequence of each characteristic parameter by a data density evaluation method; S32, constructing an index calculation submodule, wherein the index calculation submodule calculates the noise index of the normalized data sequence of each characteristic parameter according to the noise data; S33, constructing a weight assigning submodule, wherein the weight assigning submodule assigns a weight to each characteristic parameter according to the noise index.
[0006] Furthermore, the step S4 comprises: S41, constructing a noise removal submodule, wherein the noise removal submodule removes noise data from the normalized data sequence of each characteristic parameter to obtain a normalized data sequence after the noise data is removed; S42, constructing a deep learning submodule, wherein the deep learning submodule forms a feature parameter similarity matrix based on a normalized data sequence of multiple feature parameters after noise data is removed from the input; S43, constructing a risk prediction submodule, wherein the risk prediction submodule obtains risk prediction results based on a feature parameter similarity matrix.
[0007] Furthermore, the step S3 comprises: S31, constructing a noise identification submodule, wherein the noise identification submodule identifies noise data of a normalized data sequence of each characteristic parameter by a data density evaluation method; S32, constructing an index calculation submodule, wherein the index calculation submodule calculates the noise index of the normalized data sequence of each characteristic parameter according to the noise data; S33, constructing a weight assigning submodule, wherein the weight assigning submodule assigns a weight to each characteristic parameter according to the noise index.
[0008] Furthermore, the step S4 comprises: S41, constructing a noise removal submodule, wherein the noise removal submodule removes noise data from the normalized data sequence of each characteristic parameter to obtain a normalized data sequence after the noise data is removed; S42, constructing a deep learning submodule, wherein the deep learning submodule forms a feature parameter similarity matrix based on a normalized data sequence of multiple feature parameters after noise data is removed from the input; S43, constructing a risk prediction submodule, wherein the risk prediction submodule obtains risk prediction results based on a feature parameter similarity matrix.
[0009] Furthermore, in step S32, noise data statistics are performed on the normalized data sequence of each characteristic parameter I to calculate the noise index Z I : ; Where Q is the total number of noise data in the normalized data sequence of feature parameter I; f j is the number of times the jth noise data appears in the normalized data sequence of characteristic parameter I, is the position coefficient of the jth noise data in the normalized data sequence of characteristic parameter I.
[0010] Furthermore, in step S42, the normalized data sequence after removing the noise data is calculated. Each data G In The risk feature W of the category corresponding to the feature parameter I I The similarity H(In): ; The N similarities H(In) calculated from the N data constitute the similarity column vector of the feature parameter I , the similarity column vectors of M feature parameters are combined into a similarity matrix H Z =[H1…H I …H M ].
[0011] Furthermore, in step S43, the average similarity of the similarity matrix is calculated. : ; in, The weight of each feature parameter I , the average similarity and risk threshold H T A comparison is performed, and if the average similarity exceeds the risk threshold, it is judged that the individual has a risk of diabetes; if the average similarity does not exceed the risk threshold, it is judged that the individual does not have a risk of diabetes.
[0012] Furthermore, a dynamic risk threshold DH is set: DH=H T - maxD×Y -1 ×R -1 ; Where: R is a comprehensive factor considering gender and age, the male comprehensive factor is log9 (A), and the female comprehensive factor is log8 (A); Y is the dietary factor; H T is a fixed risk threshold; maxD is the maximum risk threshold counted in the big data.
[0013] Furthermore, in step S31, the density value F in the data density evaluation method is expressed as: ;
[0014] in, is the normalized data sequence representing the characteristic parameter I { }, T is the number of data in the normalized data sequence of each feature parameter, is the boundary detection radius; The optimal density value is selected to obtain the optimal boundary detection radius, and the data outside the optimal boundary detection radius is noise data.
[0015] Furthermore, in step S1, the contribution value V of the feature to the risk of disease is calculated: ;
[0016] Among them, n represents the number of groups into which the samples are divided based on this type of feature, P bi Indicates the proportion of bad samples in the i-th group of data to all samples, P yi It represents the proportion of good samples in the i-th group of data to all samples; finally, multiple important characteristic parameters with evaluation value for diabetes risk are selected whose contribution values are greater than the contribution threshold.
[0017] Compared with the prior art, the present invention has the following beneficial technical effects: A data screening unit is constructed to perform feature selection by calculating contribution values, and multiple important feature parameters with assessment value for diabetes risk are screened out. By selecting the most influential important feature parameters, the accuracy and predictive ability of the risk assessment model can be improved, the complexity of the model can be reduced, and overfitting can be avoided.
[0018] A data processing unit is constructed to obtain the monitoring data of each important characteristic parameter at multiple monitoring time points, and the monitoring data is normalized to obtain a normalized data sequence for each characteristic parameter; a parameter formation unit is constructed to calculate the noise index of the normalized data sequence of each characteristic parameter. The noise component in the data can be identified through the noise index, which helps to distinguish between signals and noise and reduce the negative impact of noise on the performance of the diabetes risk assessment model; a deep learning unit is constructed to remove the noise data in the normalized data sequence of each characteristic parameter to obtain a normalized data sequence after removing the noise data, and form a characteristic parameter similarity matrix. If the average similarity of the characteristic parameters exceeds the risk threshold, it is judged that the individual is at risk of diabetes. If the average similarity of the characteristic parameters does not exceed the risk threshold, it is judged that the individual is not at risk of diabetes. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.
[0020] Figure 1 A schematic diagram of a process for constructing a diabetes risk assessment model based on deep learning according to the present invention; Figure 2 A schematic diagram of a table of different influencing characteristic data listed in the present invention; Figure 3 It is a schematic diagram of the process of constructing a parameter forming unit of the present invention; Figure 4 A schematic diagram of a method for constructing a deep learning unit of the present invention; Figure 5 It is a structural schematic diagram of the diabetes risk assessment model based on deep learning of the present invention. DETAILED DESCRIPTION
[0021] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0022] In the drawings of the specific embodiments of the present invention, in order to better and more clearly describe the working principles of the various components in the system, the connection relationship of the various parts in the device is shown, which only clearly distinguishes the relative position relationship between the various components, and cannot constitute a limitation on the signal transmission direction, connection sequence and structural size, size and shape of each part within the component or structure.
[0023] Example 1
[0024] like Figure 1 FIG. 1 is a flow chart of a method for constructing a diabetes risk assessment model based on deep learning according to the present invention. The method for constructing a diabetes risk assessment model comprises the following steps: S1. Construct a data screening unit for feature selection by calculating contribution values, and screen out multiple important feature parameters that have assessment value for diabetes risk.
[0025] Contribution value analysis is a method to evaluate the degree of influence of different characteristics on the results. In diabetes risk assessment, this method can help identify which characteristics have a significant impact on the risk of disease.
[0026] According to the influencing characteristics of diabetes risk, different influencing characteristics data of different individuals are collected, such as Figure 2 As shown, different influencing characteristic data are listed, including: dietary structure data, age data, sleep duration data, smoking history, heart failure data, cardiovascular data, diabetic microvascular complications data, systolic blood pressure, diastolic blood pressure, cholesterol, triglycerides, creatinine, etc.
[0027] The data are divided into two samples: "good" and "bad". In diabetes risk assessment, the "good" sample refers to the characteristic data of individuals without diabetes, while the "bad" sample refers to the characteristic data of individuals with diabetes, where individuals with diabetes include individuals who develop type 2 diabetes within a preset time frame, as well as individuals with type 2 diabetes who have undergone intensive glucose-lowering therapy with a target glycated hemoglobin of less than 6.0% or canagliflozin treatment, and individuals with type 2 diabetes who have undergone standard glucose-lowering therapy with a target glycated hemoglobin of 7.0-7.9% or placebo treatment.
[0028] For each influencing feature, the ratio of "good" samples and the ratio of "bad" samples are calculated. If the ratio of "good" samples of an influencing feature is significantly higher than that of "bad" samples, it indicates that the feature has little effect on increasing the risk of diabetes. On the contrary, if the ratio of "bad" samples of an influencing feature is significantly higher than that of "good" samples, it indicates that the feature is related to increasing the risk of diabetes.
[0029] The contribution values of all influencing characteristics are comprehensively considered to determine which influencing characteristics have the greatest impact on diabetes risk, and multiple important characteristic parameters with assessment value for diabetes risk are screened.
[0030] Therefore, the present invention adopts the contribution value analysis method to measure the influence of multiple influencing features on the risk of diabetes. According to the ratio of good and bad data samples to the total data samples in the influencing feature, the contribution value V of the feature to the risk of disease is compared and calculated. The calculation formula is as follows: ; Among them, n represents the number of groups into which the samples are divided based on this type of feature, P bi It indicates the proportion of bad samples in the i-th group of data of the sample to all samples, P yi It indicates the proportion of good samples in the i-th group of data of this sample to all samples.
[0031] Finally, multiple important characteristic parameters with evaluation value for diabetes risk are selected, whose contribution values are greater than the contribution threshold.
[0032] Preferably, the multiple important characteristic parameters that have evaluation value for diabetes risk include heart failure data, cardiovascular data, diabetic microvascular complications data, systolic blood pressure data, diastolic blood pressure data, cholesterol data, etc.
[0033] S2. Construct a data processing unit to obtain monitoring data of each important characteristic parameter at multiple monitoring time points, normalize the monitoring data, and obtain a normalized data sequence of each characteristic parameter.
[0034] The monitoring data of each important characteristic parameter I at multiple monitoring time points are normalized so that the monitoring data x is mapped to the interval [-1,1] to obtain the normalized data , as shown below: , t=1,2,…T; Among them, x It is the monitoring data of the t-th monitoring time point of characteristic parameter I, T is the total number of monitoring time points, x max is the maximum data value in the monitoring data, x min is the minimum data value in the monitoring data, Represents the normalized data sequence of feature parameter I { }.
[0035] Then a normalized value set corresponding to M feature parameters is formed { }.
[0036] S3, construct a parameter forming unit, which is used to calculate the noise index of the normalized data sequence of each characteristic parameter, and assign weights to multiple characteristic parameters according to the noise index. Figure 3 As shown, the construction method of the parameter forming unit specifically includes: S31, constructing a noise identification submodule, wherein the noise identification submodule identifies the noise data of the normalized data sequence of each characteristic parameter through a data density evaluation method.
[0037] Based on the data density evaluation method, the density of the normalized data sequence of each feature parameter is evaluated, and the noise data is marked by boundary data detection.
[0038] Boundary data refers to sample points located in the sparse density area within the normalized data sequence interval of each feature parameter, which do not belong to the center range of the cluster. In the data density evaluation method, the density value F is expressed in the following form: ; in, is the normalized data sequence representing the characteristic parameter I { }, T is the number of data in the normalized data sequence of each feature parameter, It is the boundary detection radius, which controls the range of data aggregation. The boundary detection radius is crucial to data density. The window parameter determines the degree of data aggregation, thus affecting the smoothness of density estimation.
[0039] According to the empirical value, the optimal density value is selected to obtain the optimal boundary detection radius. The data outside the optimal boundary detection radius is the boundary sample, that is, the noise data.
[0040] S32, constructing an index calculation submodule, wherein the index calculation submodule calculates the noise index of the normalized data sequence of each characteristic parameter according to the noise data.
[0041] Perform noise data statistics on the normalized data sequence of each characteristic parameter I and calculate the noise index Z I : ; Where Q is the total number of noise data in the normalized data sequence of feature parameter I; f j is the number of times the jth noise data appears in the normalized data sequence of characteristic parameter I, It is the position coefficient of the jth noise data in the normalized data sequence of the characteristic parameter I. The closer the position of the noise data in the normalized data sequence of the characteristic parameter is to the density center of the sequence, the larger the position coefficient is.
[0042] S33, construct a weight assignment submodule, the weight assignment submodule is based on the noise index Z I Assign a weight to each feature parameter I , weight It is the inverse of the noise index 1 / ZI .
[0043] S4, constructing a deep learning unit, for removing noise data from the normalized data sequence of each characteristic parameter, obtaining a normalized data sequence after removing the noise data, performing deep learning based on the normalized data sequence after removing the noise data, and outputting the diabetes risk type. Figure 4 As shown, the construction method of the deep learning unit specifically includes: S41, constructing a noise removal submodule, the noise removal submodule removes noise data from the normalized data sequence of each characteristic parameter to obtain a normalized data sequence after the noise data is removed.
[0044] The noise data in the normalized data sequence of each characteristic parameter is removed to obtain the normalized data sequence after the noise data is removed. ,N=TQ.
[0045] S42. Construct a deep learning submodule, wherein the deep learning submodule forms a feature parameter similarity matrix based on a normalized data sequence of multiple input feature parameters after noise data is removed.
[0046] The deep learning submodule first calculates the normalized data sequence after removing the noise data Each data G In The risk feature W of the category corresponding to the feature parameter I I The similarity H(In): ; Get the similarity column vector of feature parameter I , the similarity column vectors of M feature parameters I are combined into a similarity matrix H Z =[H1…H I …H M ].
[0047] S43, constructing a risk prediction submodule, wherein the risk prediction submodule outputs a risk prediction result based on a feature parameter similarity matrix.
[0048] Calculate the average similarity of the feature parameter similarity matrix , ; The weight of each feature parameter I , the average similarity calculated and risk threshold H T Compare and if the average similarity exceeds the risk threshold H T , then the individual is judged to have diabetes risk. If the average similarity does not exceed the risk threshold H T, then the individual is judged to be not at risk of diabetes. It should be emphasized that the risk threshold is set based on previous studies, statistical analysis or expert opinions to distinguish whether there is a risk of diabetes.
[0049] Example 2 On the basis of Example 1, for the step of constructing a risk prediction submodule, based on the feature parameter similarity matrix, outputting the technical solution of the risk prediction result, this embodiment adopts a method of setting a dynamic threshold according to the age, diet, and gender of the individual to be predicted to set the risk threshold. The specific method is as follows: The calculation formula of dynamic threshold DH is: DH=H T - maxD×Y -1 ×R -1 ; Wherein: R is a comprehensive factor that takes gender and age into consideration. Specifically, the male factor R1 is log9(A), that is, the female factor R2 is log8(A); that is, the value of the logarithm of the real number with 9 or 8 as the base and age A as the logarithm.
[0050] Y is the dietary factor, which is used to obtain the dietary structure of the individual to be predicted. The worse the dietary structure, the smaller the value of the dietary factor I. T is a fixed risk threshold; maxD is the maximum risk threshold counted in the big data.
[0051] Example 3
[0052] The structure of the diabetes risk assessment model constructed using the above-mentioned diabetes risk assessment model construction method based on deep learning is as follows: Figure 5 As shown, the diabetes risk assessment model structure includes: a data screening unit, a data processing unit, a parameter formation unit and a deep learning unit.
[0053] According to the influencing characteristics of diabetes risk, different influencing characteristic data of different individuals are collected and input into the data screening unit. The data screening unit performs feature selection by calculating the contribution value, and screens out multiple important characteristic parameters that have assessment value for diabetes risk, such as: systolic blood pressure, diastolic blood pressure, cholesterol, triglycerides, creatinine, etc.
[0054] Obtain monitoring data for each important characteristic parameter at multiple monitoring time points. These monitoring data can come from sensors, inspection instruments or other data sources, and ensure that each important characteristic parameter has corresponding monitoring values at all monitoring time points to avoid missing values. If there are missing values in the monitoring data, you can use interpolation (such as linear interpolation, polynomial interpolation, etc.) or fill in according to business logic (such as using mean, median, mode, etc.). If the monitoring data contains timestamps, you need to ensure that the timestamp format is unified.
[0055] The monitoring data is input into the data processing unit, and the data processing unit performs normalization processing on the monitoring data to obtain a normalized data sequence of each characteristic parameter, thereby forming a normalized value set corresponding to multiple characteristic parameters.
[0056] The normalized data sequence of the characteristic parameters is input into the parameter forming unit in sequence, and the parameter forming unit calculates the noise index of the normalized data sequence of each characteristic parameter, and assigns weights to the plurality of characteristic parameters according to the noise index.
[0057] Specifically, the parameter forming unit includes: a noise identification submodule, an index calculation submodule and a weight assignment submodule. The noise identification submodule selects the optimal density value through the data density evaluation method to obtain the optimal boundary detection radius. The data outside the optimal boundary detection radius is the boundary sample, that is, the noise data of the normalized data sequence of each characteristic parameter is identified; the index calculation submodule calculates the noise index of the normalized data sequence of each characteristic parameter according to the noise data; the weight assignment submodule assigns a weight to each characteristic parameter according to the noise index, and the weight is the inverse of the noise index.
[0058] The deep learning unit removes noise data from the normalized data sequence of each characteristic parameter to obtain a normalized data sequence after the noise data is removed, and performs diabetes risk assessment based on the normalized data sequence after the noise data is removed.
[0059] Specifically, the deep learning unit includes: a noise removal submodule, a deep learning submodule and a risk prediction submodule.
[0060] The noise removal submodule removes the noise data in the normalized data sequence of each characteristic parameter to obtain a normalized data sequence after the noise data is removed.
[0061] The architecture of the deep learning submodule includes a shared layer for learning common features and a specific task layer for row refinement prediction. The shared layer is composed of a convolutional layer, a recurrent layer or a fully connected layer, and is responsible for learning and analyzing from all data. Specifically, in the present invention, the shared layer calculates the similarity between each data in the normalized data sequence after removing the noise data and the risk feature of the category corresponding to the feature parameter. Before the output of the shared layer is input to the specific task layer, it needs to be sliced according to multiple feature parameters to ensure that the data of each feature parameter is processed separately. The similarity column vectors of the output multiple feature parameters are combined into a similarity matrix and used as the input of the specific task layer; for each individual to be predicted, there is a series of linear layer combinations, and these linear layers set dynamic thresholds according to the age, diet, and gender of the individual to be predicted received from the shared layer to generate prediction results that are consistent with the age stage of the individual to be predicted. Specifically, the specific task layer compares the calculated average similarity with the risk threshold. If the average similarity exceeds the risk threshold, it is judged that the individual has a risk of diabetes; if the average similarity does not exceed the risk threshold, it is judged that the individual does not have a risk of diabetes.
[0062] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media integrated. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid state disk (SSD)), etc.
[0063] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.
Claims
1. A method for constructing a diabetes risk assessment model based on deep learning, characterized in that: The steps include: S1. Constructing a data screening unit, wherein the data screening unit performs feature selection by calculating contribution values, and screens out a plurality of important feature parameters that have evaluation value for diabetes risk; S2. Constructing a data processing unit, wherein the data screening unit obtains monitoring data of each important characteristic parameter at multiple monitoring time points, normalizes the monitoring data, and obtains a normalized data sequence of each characteristic parameter; S3, constructing a parameter forming unit, wherein the parameter forming unit calculates a noise index of a normalized data sequence of each characteristic parameter, and assigns weights to multiple characteristic parameters according to the noise index; S4. Construct a deep learning unit, wherein the deep learning unit removes noise data from the normalized data sequence of each characteristic parameter to obtain a normalized data sequence after the noise data is removed, performs deep learning based on the normalized data sequence after the noise data is removed, and outputs a diabetes risk type.
2. The method for constructing a diabetes risk assessment model according to claim 1, characterized in that: In step S2, the monitoring data of each important characteristic parameter I at multiple monitoring time points are normalized so that the monitoring data is mapped to the interval [-1, 1] to obtain normalized data : ,t=1,2,……T; Among them, x It is the monitoring data of the t-th monitoring time point of characteristic parameter I, T is the total number of monitoring time points, x max is the maximum data value in the monitoring data, x min is the minimum data value in the monitoring data, Represents the normalized data sequence of feature parameter I { }.
3. The method for constructing a diabetes risk assessment model according to claim 1, characterized in that: The step S3 comprises: S31, constructing a noise identification submodule, wherein the noise identification submodule identifies noise data of a normalized data sequence of each characteristic parameter by a data density evaluation method; S32, constructing an index calculation submodule, wherein the index calculation submodule calculates the noise index of the normalized data sequence of each characteristic parameter according to the noise data; S33, constructing a weight assigning submodule, wherein the weight assigning submodule assigns a weight to each characteristic parameter according to the noise index.
4. The method for constructing a diabetes risk assessment model according to claim 1, characterized in that: The step S4 comprises: S41, constructing a noise removal submodule, wherein the noise removal submodule removes noise data from the normalized data sequence of each characteristic parameter to obtain a normalized data sequence after the noise data is removed; S42, constructing a deep learning submodule, wherein the deep learning submodule forms a feature parameter similarity matrix based on a normalized data sequence of multiple feature parameters after noise data is removed from the input; S43, constructing a risk prediction submodule, wherein the risk prediction submodule obtains risk prediction results based on a feature parameter similarity matrix.
5. The method for constructing a diabetes risk assessment model according to claim 3, characterized in that: In step S32, noise data statistics are performed on the normalized data sequence of each characteristic parameter I to calculate the noise index Z I : ； Where Q is the total number of noise data in the normalized data sequence of feature parameter I; f j is the number of times the jth noise data appears in the normalized data sequence of characteristic parameter I, is the position coefficient of the jth noise data in the normalized data sequence of characteristic parameter I.
6. The method for constructing a diabetes risk assessment model according to claim 4, characterized in that: In step S42, the normalized data sequence after removing the noise data is calculated. Each data G In The risk feature W of the category corresponding to the feature parameter I I The similarity H(In): ； The N similarities H(In) calculated from the N data constitute the similarity column vector of the feature parameter I , the similarity column vectors of M feature parameters are combined into a similarity matrix H Z =[H1…H I …H M ].
7. The method for constructing a diabetes risk assessment model according to claim 6, characterized in that: In step S43, the average similarity of the similarity matrix is calculated. : ； in, The weight of each feature parameter I , the average similarity and risk threshold H T A comparison is performed, and if the average similarity exceeds the risk threshold, it is judged that the individual has a risk of diabetes; if the average similarity does not exceed the risk threshold, it is judged that the individual does not have a risk of diabetes.
8. The method for constructing a diabetes risk assessment model according to claim 7, characterized in that: Setting the dynamic risk threshold DH: DH=H T - maxD×Y -1 ×R -1 ; Where: R is a comprehensive factor considering gender and age, the comprehensive factor for men is log9 (A), and the comprehensive factor for women is log8 (A); Y is the dietary factor; H T is a fixed risk threshold; maxD is the maximum risk threshold counted in the big data.
9. The method for constructing a diabetes risk assessment model according to claim 3, characterized in that: In step S31, the density value F in the data density evaluation method is expressed as: ； in, is the normalized data sequence representing the characteristic parameter I { }, T is the number of data in the normalized data sequence of each feature parameter, is the boundary detection radius; The optimal density value is selected to obtain the optimal boundary detection radius, and the data outside the optimal boundary detection radius is noise data.
10. The method for constructing a diabetes risk assessment model according to claim 1, characterized in that: In step S1, the contribution value V of the feature to the risk of disease is calculated: ； Among them, n represents the number of groups into which the samples are divided based on this type of feature, P bi Indicates the proportion of bad samples in the i-th group of data to all samples, P yi It represents the proportion of good samples in the i-th group of data to all samples; finally, multiple important characteristic parameters with contribution values greater than the contribution threshold that are valuable for assessing diabetes risk are selected.