Dynamic classification method and device based on improved ensemble strategy for disease data
Through the combination of fuzzy clustering and dynamic selective subclassifier pool, the problem of insufficient classification accuracy of disease data is solved, especially in the case of data imbalance, and higher classification accuracy and multi-angle evaluation ability are achieved.
Patent Information
- Application Number
- PCT/CN2023/097026
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-23
- Filing Date
- 2023-05-30
- Publication Date
- 2024-11-28
AI Technical Summary
Existing disease data classification methods are insufficient in the face of diverse and complex disease data, especially due to poor classification results due to data imbalance.
Adopting an improved integration strategy for disease data, through the combination of fuzzy clustering, dynamic selection subclassifier pool and test sample weights, the decision threshold is adjusted in combination with the category ratio of the data set to form a new integration model to improve classification accuracy.
It significantly improves the classification accuracy of disease data, can better handle complex and unbalanced data sets, provide more accurate classification results, and evaluate the accuracy of prediction results through multiple angles.
Smart Images

Figure CN2023097026_28112024_PF_FP_ABST
Abstract
Description
A dynamic classification method and device for improving integration strategy of disease data Technical Field
[0001] The present invention relates to the field of data processing technology, and in particular to a dynamic classification method and device for improving disease data integration strategy. Background Art
[0002] Traditional methods for disease prediction and diagnosis generally rely on physician experience, a process that requires extensive accumulation, and external environmental interference can also impact physician predictions. With the continuous advancement of machine learning, scholars have been attempting to classify the various attributes of disease data in recent years, hoping that the classification results can assist in medical purposes such as disease prediction and diagnosis.
[0003] While data classification models based on single classifiers can achieve their objectives to a certain extent, the wide variety of diseases and the increasing complexity and diversity of disease data make single-classifier-based classification models no longer guarantee accurate predictions. This has led to the emergence of ensemble learning, which combines multiple weak learners into a single strong learner, combining the strengths of each base classifier and offering improved performance compared to single classifiers. However, as the number of subclassifiers increases, the differences between them decrease. While selective ensembles can further improve the effectiveness and generalization of ensemble learning methods, they also suffer from high computational complexity. Furthermore, centralized disease data suffers from severe data imbalance, making conventional selective ensemble methods incapable of achieving ideal classification results.
[0004] Summary of the Invention
[0005] In response to the above problems, the present invention provides a dynamic classification method and device for improving the integration strategy of disease data, which effectively solves the problem of poor classification effect of existing methods.
[0006] The technical solutions provided by the present invention are as follows:
[0007] In one aspect, the present invention provides a dynamic classification method for improving integration strategies for disease data, comprising:
[0008] Obtain the disease data to be classified and split it into training set and test set according to the preset ratio;
[0009] Perform fuzzy clustering on the disease data contained in the training set into two categories of samples, and find a preset number of neighboring samples of the two cluster centers as validation samples;
[0010] Inputting the verification samples into a pre-configured weak classifier set containing different types of classifiers, and selecting a preset number of classifiers from the weak classifier set according to the classification results to form a sub-classifier pool;
[0011] For a test sample in the test set, its weight in each category is obtained according to the cluster center, and the test sample is input into the sub-classifier pool to obtain its predicted probability in different categories;
[0012] Combining the weights and predicted probabilities, we can get the final probabilities of the test sample in different categories.
[0013] Determining a decision threshold for each sub-classifier based on the final probability and the number of samples in each category after the training set is classified;
[0014] The test samples are finally classified according to the determined decision threshold.
[0015] In another aspect, the present invention provides a dynamic classification device for improving an integration strategy for disease data, comprising:
[0016] The data acquisition module is used to obtain the disease data to be classified and split it into training set and test set according to the preset ratio;
[0017] A fuzzy clustering module is used to fuzzy cluster the disease data contained in the training set into two types of samples, and find a preset number of neighboring samples of the two cluster centers as verification samples;
[0018] a sub-classifier pool forming module, configured to input the verification samples into a pre-configured weak classifier set containing different types of classifiers, and select a preset number of classifiers from the weak classifier set according to the classification results to form a sub-classifier pool;
[0019] a classification probability determination module for determining, for a test sample in the test set, its weight for each category based on the cluster center, inputting the test sample into the sub-classifier pool to obtain its predicted probability for different categories; and combining the weight and predicted probability to obtain the final probability of the test sample for different categories;
[0020] A decision threshold determination module, configured to determine a decision threshold for each sub-classifier based on the final probability and the number of samples in each category after the training set is classified;
[0021] The classification module is used to perform the final classification of the test samples according to the determined decision threshold.
[0022] On the other hand, the present invention provides a terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned dynamic classification method steps when executing the computer program.
[0023] On the other hand, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned dynamic classification method are implemented.
[0024] The dynamic classification method and device for improving disease data integration strategy provided by the present invention can at least bring the following beneficial effects:
[0025] 1. We combined the fuzzy clustering algorithm, the dynamic selection of subclassifier pools, the idea of dynamic weighting of test samples, and the method of changing the decision threshold to address data imbalance to obtain a new integrated model. This model is learned based on existing disease data to obtain a classification model, which is used to classify sample data, greatly improving classification accuracy.
[0026] 2. Select multiple weak classifiers as a sub-classifier pool. Dynamically select sub-classifiers based on their validation results on different datasets. Combining the strengths of multiple classifiers, the resulting ensemble classifier has stronger classification capabilities and can better complete the corresponding classification task. Even with increasingly complex medical disease data, more accurate classification results can be obtained.
[0027] 3. To address data imbalance in disease datasets, we improve prediction accuracy by adjusting the decision threshold in the final decision stage based on the dataset's category ratio. Each test sample has its own calculation weight, resulting in better classification results compared to the fixed weight calculation used in traditional classification models.
[0028] 4. Evaluate prediction results from multiple indicators and angles. For the obtained prediction results, multiple indicators are used to evaluate them from different angles to better understand the accuracy of the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] The preferred implementation scheme will be described below in a clear and understandable manner with reference to the accompanying drawings to further illustrate the above characteristics, technical features, advantages and their implementation methods.
[0030] FIG1 is a flow chart of a dynamic classification method for improving disease data integration strategy according to the present invention.
[0031] FIG2 is a schematic diagram of the structure of a dynamic classification device for improving the integration strategy of disease data according to the present invention.
[0032] FIG3 is a schematic diagram of the terminal device structure of the present invention.
[0033] Reference numerals: 100 - dynamic classification device, 110 - data acquisition module, 120 - fuzzy clustering module, 130 - subclassifier pool formation module, 140 - classification probability determination module, 150 - decision threshold determination module, 160 - classification module. DETAILED DESCRIPTION
[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the specific embodiments of the present invention will be described below with reference to the accompanying drawings. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings and other embodiments can be obtained based on these drawings without inventive work.
[0035] One embodiment of the present invention provides a dynamic classification method for improving disease data integration strategy, as shown in FIG1 , including:
[0036] S10 obtains the disease data to be classified and splits it into a training set and a test set according to a preset ratio.
[0037] The acquired disease data to be classified belongs to the same type of data, such as disease data collected by a hospital for the same disease (lung cancer, diabetes, etc.), or data stored in a database for the same disease. The division of the training set and test set can be determined based on actual conditions. Generally speaking, 80% of the disease data is divided into the training set for training the classification model, and 20% of the disease data is divided into the test set for testing the trained classification model.
[0038] In other embodiments, in order to improve the accuracy of data classification, after obtaining the data set of disease data to be classified, data cleaning should be performed on it, including processing missing values and outliers in the data set, and for data containing various attributes in the data set, the data should be converted to data at the same level through mean-variance normalization:
[0039] Among them, x * represents the normalized value, x represents the data in the dataset, represents the mean of the data in the dataset, and σ represents the standard deviation of the data in the dataset.
[0040] S20 fuzzy clusters the disease data contained in the training set into two categories of samples, and finds a preset number of neighboring samples of the two cluster centers as verification samples.
[0041] In this embodiment, the problem of binary classification (1 category or 0 category) of disease data is targeted. The classification attributes targeted in the classification can be selected according to the actual situation and are not specifically limited here. For example, when the classification goal is to determine whether indicator A exceeds the threshold, a fuzzy clustering algorithm such as FCM is used to cluster it, and the training set is divided into two categories: sample S0 (samples whose indicator A does not exceed the threshold) and S1 (samples whose indicator A exceeds the threshold). After obtaining the cluster centers C0 and C1 of the two categories, N neighboring samples of the cluster centers are found as verification sample sets V0 and V1. The number of neighboring samples is also determined according to the actual application and is not specifically limited here.
[0042] S30 inputs the verification samples into a pre-configured weak classifier set containing different types of classifiers, and selects a preset number of classifiers from the weak classifier set according to the classification results to form a sub-classifier pool.
[0043] Weak classifier sets include various classifiers, such as KNN (nearest neighbor algorithm), ID3 (a decision tree algorithm), C4.5 (a decision tree algorithm), artificial neural networks (ANN), Fisher discriminant analysis (FDA), naive Bayes (NB), decision trees (DT), support vector machines (SVM), logistic regression (LR), and random forests (RF). The selection of classifiers in a weak classifier set depends on the specific application. To improve classification results, it's recommended to include all commonly used classifiers in the application. Of course, the more types included, the longer the time required, so appropriate selection is recommended.
[0044] After the verification sample is input into the weak classifier set (the same sample data is input into each classifier), according to the level of the classification result F1-score of each classifier, a preset number of classifiers with higher classification result F1-score are selected to form a sub-classifier pool. Here, the number of sub-classifiers included in the sub-classifier pool is also set according to the actual application. As in an example, 3, 5, 7 or even more classifiers with the highest classification result F1-score are selected to form a sub-classifier pool. In this process, the parameters of each classifier in the weak classifier set are selected by default, and of course specific parameters can also be set according to demand.
[0045] The selected sub-classifier pool combines the advantages of multiple sub-classifiers, which can better complete the corresponding classification tasks and improve the classification accuracy.
[0046] S40 obtains the weight of a test sample in the test set on each category according to the cluster center, inputs the test sample into the sub-classifier pool, and obtains its predicted probability on different categories.
[0047] The subclassifier pool includes multiple subclassifiers. In this step, the weights of different categories (2 categories) in each subclassifier are calculated separately. For a subclassifier:
[0048] The i-th test sample (x i1 ,x i2 ,…,x iF ) and the k-th cluster center vector (x k1 ,x k2 ,…,x kF ) weight W ik for:
[0049] Where F represents the characteristic dimension of the test samples in the test set, which is determined by the acquired disease data. Different types of disease data have different characteristic dimensions; x if represents the fth attribute (feature) of the i-th test sample; c kf Represents the fth attribute (feature) of the kth class center (center sample).
[0050] Since this embodiment solves a binary classification problem, the classification result includes two categories, corresponding to two class centers. Specifically, k=0 represents category 0, and k=1 represents category 1:
[0051] The weight W between the i-th test sample and the center vector of category 0 i0 for:
[0052] The weight W between the i-th test sample and the center vector of category 1 i1 for:
[0053] Among them, c 0f represents the fth attribute of the center sample under category 0; c 1f Represents the fth attribute of the center sample under category 1.
[0054] S50 combines the weights and predicted probabilities to obtain final probabilities of the test sample in different categories.
[0055] After calculating the weights for the two categories of each sub-classifier, the weights and predicted probabilities are further combined to obtain the final probability of the test sample in different categories in each sub-classifier. This process is also calculated for each sub-classifier. The calculation process for a sub-classifier is as follows:
[0056] S51 is based on the weights W of different categories in the sub-classifier ik Get the sample weight matrix W of the test sample iK : W iK =[W i0,W i1 ]
[0057] Where K represents the number of categories, K = 2; W i0 represents the weight of category 0 of the i-th test sample, W i1 represents the weight of category 1 of the i-th test sample;
[0058] S52 is based on the obtained sample weight matrix W iK And the predicted probability is used to obtain the final probability P on the corresponding sub-classifier ik : P ik =W iK *p ik
[0059] Among them, p ik It represents the predicted probability of the i-th test sample in the k-th classification category, specifically:
[0060] When k=0, the final probability P of the 0 class i0 =W iK *p i0
[0061] When k=1, the final probability P of the 1 class i1 =W iK *p i1 .
[0062] In this process, a unique weight is configured for each test sample, which can improve the classification effect to a certain extent compared with the fixed weight calculation of the traditional classification model.
[0063] S60 determines the decision threshold of each sub-classifier based on the final probability and the number of samples in each category after the training set is classified.
[0064] Based on step S50, each test sample will obtain the final probability P1 and P0 on each sub-classifier. In this way, the decision threshold of each sub-classifier is further determined according to the final probability in step S60:
[0065] S61 performs fuzzy clustering on the disease data in the training set to obtain the number of samples N1 and N0 in two categories.
[0066] The training set is fuzzy clustered into two categories of samples. The two categories include a certain number of samples. The number of samples contained in category 1 is defined as N1, and the number of samples contained in category 0 is defined as N0.
[0067] S62 calculates the ratio of the number of samples of the two categories I1=N1 / N0 and the final probability ratio of the i-th test sample in the two categories in the classifier I2=P i1 / P i0 ;
[0068] S63 determines the size of the sample number ratio I1 and the final probability ratio I2; when the final probability ratio I2 is greater than the sample number ratio I1 (I2>I1), the test sample is determined to be category 1, otherwise the test sample is determined to be category 0.
[0069] It should be understood that in other embodiments, when the sample number ratio I1′=N0 / N1 in step S62, the corresponding final probability ratio I2′=P i0 / P i1 , then when the final probability ratio I2' is not greater than the sample number ratio I1', the test sample is determined to be category 1, otherwise the test sample is determined to be category 0.
[0070] According to the above decision threshold determination method, due to the different weights of each sub-classifier, the final decision threshold is also different. In practical applications, disease datasets are often imbalanced, and direct classification results will be biased towards the majority class. Using the method in this embodiment, the decision threshold is adjusted according to the sample ratio of different datasets, increasing the proportion of minority class samples, which can greatly improve classification accuracy.
[0071] S70 performs a final classification on the test sample according to the determined decision threshold.
[0072] The above embodiment is improved. In this embodiment, after each sub-classifier performs a final classification on the test sample according to the determined decision threshold in step S70, the following steps are further included:
[0073] S80 evaluates the subclassifier pool based on pre-configured evaluation metrics, including CE:
[0074] Where N represents the total number of samples of disease data in the test set; K represents the number of categories, K = 2; a + Indicates that weight is given The sensitivity TPR (True Position Rate) indicates the classification accuracy of 1 class of samples; N1 indicates the number of samples classified into 1 class in the test set; a - Indicates that weight is given The True Negative Rate (TNR) represents the classification accuracy of class 0 samples; N0 represents the number of samples in the test set classified as class 0, TP represents the number of samples correctly predicted as class 1, FN represents the number of samples incorrectly predicted as class 0, TN represents the number of samples correctly predicted as class 0, and TN represents the number of samples incorrectly predicted as class 1. A larger CE indicator indicates a better classification performance of the subclassifier pool.
[0075] In addition to the CE indicator, in other embodiments, the accuracy of the classification results is evaluated from multiple perspectives based on multiple different indicators, including accuracy, precision, recall, F1-score, AUC, sensitivity, TPR, and specificity, TNR, to verify the classification prediction ability of the model. Specifically:
[0076] In this embodiment, for the obtained classification results, the sample imbalance problem of the disease data set is taken into consideration, new weights are assigned according to the actual number of sample categories, and a new evaluation index CE is defined. At the same time, the accuracy of the results is evaluated from different angles through multiple evaluation indicators such as accuracy, precision, recall rate, F1 score, AUC, sensitivity, and specificity.
[0077] The present invention also provides a dynamic classification device 100 for improving the integration strategy of disease data, comprising: a data acquisition module 110 for acquiring disease data to be classified and splitting it into a training set and a test set according to a preset ratio; a fuzzy clustering module 120 for fuzzy clustering the disease data contained in the training set into two types of samples, and finding a preset number of neighboring samples of the two cluster centers as verification samples; a subclassifier pool forming module 130 for inputting the verification samples into a pre-configured weak classifier set containing different types of classifiers, and selecting a pre-classified weak classifier from the weak classifier set according to the classification results. Assume that a number of classifiers form a sub-classifier pool; a classification probability determination module 140 is used to obtain the weight of a test sample in the test set in each category according to the cluster center, input the test sample into the sub-classifier pool, and obtain its predicted probability in different categories; and is used to combine the weight and the predicted probability to obtain the final probability of the test sample in different categories; a decision threshold determination module 150 is used to determine the decision threshold of each sub-classifier based on the final probability and the number of samples in each category after the training set is classified; a classification module 160 is used to finally classify the test sample according to the determined decision threshold.
[0078] In this embodiment, the disease data to be classified acquired by the data acquisition module 110 is of the same type, such as disease data collected by a hospital for the same disease (lung cancer, diabetes, etc.), or data for the same disease stored in a database. The division of the training set and the test set can be determined based on actual conditions. Generally speaking, 80% of the disease data is divided into a training set for training the classification model, and 20% of the disease data is divided into a test set for testing the trained classification model. In other embodiments, the dynamic classification device 100 also includes a data cleaning module. To improve the accuracy of data classification, after the data acquisition module 110 acquires the data set of the disease data to be classified, it uses the data cleaning module to clean the data. This includes processing missing values and outliers in the data set, and converting the data containing various attributes in the data set to the same level through mean-variance normalization.
[0079] Since the problem is a binary classification (1 category or 0 category) of disease data, the classification attributes targeted in the classification can be selected according to the actual situation and are not specifically limited here. For example, when the classification goal is to determine whether indicator A exceeds the threshold, the fuzzy clustering module 120 uses a fuzzy clustering algorithm such as FCM to cluster it, and divides the training set into two categories: samples S0 (samples whose indicator A does not exceed the threshold) and S1 (samples whose indicator A exceeds the threshold). After obtaining the cluster centers C0 and C1 of the two categories, N neighboring samples of the cluster centers are found as verification sample sets V0 and V1. The number of neighbor samples is also determined according to the actual application and is not specifically limited here. In actual applications, it is also possible to make an early prediction of whether the disease data is suffering from cancer. At this time, category 1 corresponds to cancer and category 0 corresponds to health.
[0080] Weak classifier sets include various classifiers, such as KNN (nearest neighbor algorithm), ID3 (a decision tree algorithm), C4.5 (a decision tree algorithm), artificial neural networks (ANN), Fisher discriminant analysis (FDA), naive Bayes (NB), decision trees (DT), support vector machines (SVM), logistic regression (LR), and random forests (RF). The selection of classifiers in a weak classifier set depends on the specific application. To improve classification results, it's recommended to include all commonly used classifiers in the application. Of course, the more types included, the longer the time required, so appropriate selection is recommended.
[0081] After the subclassifier pool forming module 130 inputs the verification sample into the weak classifier set (the same sample data is input into each classifier), according to the level of the classification result F1-score of each classifier, a preset number of classifiers with higher classification result F1-score are selected to form the subclassifier pool. Here, the number of subclassifiers included in the subclassifier pool is set according to actual application. As in an example, 3, 5, 7 or even more classifiers with the highest classification result F1-score are selected to form the subclassifier pool. In this process, the parameters of each classifier in the weak classifier set are selected by default, and of course specific parameters can also be set according to demand.
[0082] Since the sub-classifier pool includes multiple sub-classifiers, the classification probability determination module 140 will calculate the weights of different categories (2 categories) in each sub-classifier separately. For a sub-classifier, the i-th test sample (x i1 ,x i2 ,…,x iF ) and the k-th cluster center vector (x k1 ,x k2 ,…,x kF ) weight W ik for:
[0083] Where F represents the characteristic dimension of the test samples in the test set, which is determined by the acquired disease data. Different types of disease data have different characteristic dimensions; x if represents the fth attribute (feature) of the i-th test sample; c kf Represents the fth attribute (feature) of the kth class center (center sample).
[0084] After respectively calculating the weights for the two classifications of each sub-classifier, the classification probability determination module 140 further combines the weights and the predicted probabilities to obtain the final probabilities of the test sample in different classifications in each sub-classifier.
[0085] Specifically, the classification probability determination module 140 includes:
[0086] The weight matrix determination unit is used to determine the weights W of different categories in a sub-classifier. ik Get the sample weight matrix W of the test sample iK : W iK =[W i0 ,W i1 ]
[0087] Where K represents the number of categories, K = 2; W i0 represents the weight of category 0 of the i-th test sample, W i1 represents the weight of category 1 of the i-th test sample;
[0088] The final probability determination unit is used to determine the sample weight matrix W according to the obtained iK And the predicted probability is used to obtain the final probability P on the corresponding sub-classifier ik : P ik =W iK *p ik
[0089] Among them, p ik It represents the final probability of the i-th test sample in the k-th classification category.
[0090] The decision threshold determination module 150 includes:
[0091] The sample number determination unit is used to perform fuzzy clustering on the disease data in the training set to obtain the sample numbers N1 and N0 of the two categories;
[0092] The ratio determination unit is used to calculate the ratio of the number of samples of the two categories I1=N1 / N0 and the final probability ratio I2=P of the i-th test sample on the two categories in the classifier for a sub-classifier. i1 / P i0 ;
[0093] The decision threshold judgment unit is used to judge the size of the sample number ratio I1 and the final probability ratio I2; when the final probability ratio I2 is greater than the sample number ratio I1, the test sample is judged to be category 1, otherwise the test sample is judged to be category 0.
[0094] The present embodiment is obtained by improving the above embodiment. In the present embodiment, the dynamic classification device 100 further includes an evaluation module for evaluating the sub-classifier pool according to pre-configured evaluation indicators. The evaluation indicators include indicator CE:
[0095] Where N is the total number of samples of disease data in the training set; K is the number of categories, K = 2; a + Indicates that weight is given The sensitivity indicates the classification accuracy of class 1 samples, and N1 indicates the number of samples classified into class 1 in the training set; a - Indicates that weight is given Specificity, N0 represents the number of samples classified as category 0 in the training set, TP represents the number of samples correctly predicted as category 1, FN represents the number of samples incorrectly predicted as category 0, TN represents the number of samples correctly predicted as category 0, and TN represents the number of samples incorrectly predicted as category 1.
[0096] In addition to the CE indicator, in other embodiments, the accuracy of the classification results is evaluated from multiple perspectives based on multiple different indicators, including accuracy, precision, recall, F1-score, AUC, sensitivity, TPR, and specificity, TNR, to verify the classification prediction ability of the model. Specifically:
[0097] In this embodiment, for the obtained classification results, the sample imbalance problem of the disease data set is taken into consideration, new weights are assigned according to the actual number of sample categories, and a new evaluation index CE is defined. At the same time, the accuracy of the results is evaluated from different angles through multiple evaluation indicators such as accuracy, precision, recall rate, F1 score, AUC, sensitivity, and specificity.
[0098] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned program modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different program modules as needed, that is, the internal structure of the device can be divided into different program units or modules to complete all or part of the functions described above. The program modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one processing unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software program unit. In addition, the specific names of the program modules are only for the convenience of distinguishing each other and are not used to limit the scope of protection of the present invention.
[0099] FIG3 is a schematic diagram of the structure of a terminal device provided in one embodiment of the present invention. As shown, the terminal device 200 includes a processor 220, a memory 210, and a computer program 211 stored in the memory 210 and executable on the processor 220, such as a dynamic classification program. When the processor 220 executes the computer program 211, the steps of the aforementioned dynamic classification embodiments are implemented. Alternatively, when the processor 220 executes the computer program 211, the functions of the modules of the aforementioned dynamic classification apparatus embodiments are implemented.
[0100] The terminal device 200 may be a notebook, a PDA, a tablet computer, a mobile phone, or other device. The terminal device 200 may include, but is not limited to, a processor 220 and a memory 210. Those skilled in the art will appreciate that FIG3 is merely an example of the terminal device 200 and does not limit the terminal device 200. The terminal device 200 may include more or fewer components than shown, or may combine certain components or different components. For example, the terminal device 200 may also include input / output devices, a display device, a network access device, a bus, and the like.
[0101] The processor 220 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor 220 may be a microprocessor or any conventional processor.
[0102] The memory 210 can be an internal storage unit of the terminal device 200, such as a hard disk or memory of the terminal device 200. The memory 210 can also be an external storage device of the terminal device 200, such as a plug-in hard disk equipped on the terminal device 200, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. Furthermore, the memory 210 can also include both an internal storage unit of the terminal device 200 and an external storage device. The memory 210 is used to store a computer program 211 and other programs and data required by the terminal device 200. The memory 210 can also be used to temporarily store data that has been output or is about to be output.
[0103] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.
[0104] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.
[0105] In the embodiments provided by the present invention, it should be understood that the disclosed terminal devices and methods can be implemented in other ways. For example, the terminal device embodiments described above are merely illustrative. For example, the division of modules or units is merely a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.
[0106] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.
[0107] In addition, the functional units in various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.
[0108] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by sending instructions to the relevant hardware through the computer program 211. The computer program 211 can be stored in a computer-readable storage medium. When the computer program 211 is executed by the processor 220, it can implement the steps of the above-mentioned various method embodiments. Among them, the computer program 211 includes: computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device that can carry the computer program 211 code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in computer-readable storage media can be appropriately increased or decreased according to the requirements of legislation and patent practices in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practices, computer-readable media do not include electrical carrier signals and telecommunications signals.
[0109] It should be noted that the above embodiments can be freely combined as needed. The above are only preferred embodiments of the present invention. It should be pointed out that ordinary relevant personnel in this technical field can make several improvements and modifications without departing from the principles of the present invention. Such improvements and modifications should also be considered as the scope of protection of the present invention.
Claims
1. A dynamic classification method for improving integration strategy for disease data, It is characterized in that include: Obtain the disease data to be classified and split it into training set and test set according to the preset ratio; Perform fuzzy clustering on the disease data contained in the training set into two types of samples, and find a preset number of neighboring samples of the two cluster centers as verification samples; Inputting the verification samples into a pre-configured weak classifier set containing different types of classifiers, respectively, and selecting a preset number of classifiers from the weak classifier set according to the classification results to form a sub-classifier pool; For a test sample in the test set, obtain its weight in each category according to the cluster center, input the test sample into the sub-classifier pool, and obtain its predicted probability in different categories; Combining the weights and predicted probabilities, the final probabilities of the test samples in different categories are obtained; Determining a decision threshold of each sub-classifier based on the final probability and the number of samples in each category after the training set is classified; The test samples are finally classified according to the determined decision threshold.
2. The dynamic classification method according to claim 1, It is characterized in that The weights of each category obtained according to the cluster center are as follows: ik for: Among them, F represents the characteristic dimension of the test samples in the test set; x if represents the fth attribute of the i-th test sample; c kf represents the fth attribute of the kth class center, k=0 represents category 0, and k=1 represents category 1.
3. The dynamic classification method according to claim 2, It is characterized in that Combining the weights and predicted probabilities, the final probabilities of the test sample in different categories are obtained, including: According to the weights W of different categories in a sub-classifier ik Get the sample weight matrix W of the test sample iK : IN iK =[W i0 ,IN i1 ] Where K represents the number of categories, K = 2; W i0 represents the weight of the 0 category of the i-th test sample, W i1 represents the weight of category 1 of the i-th test sample; According to the obtained sample weight matrix W iK And the predicted probability obtains the final probability P on the corresponding sub-classifier ik : P ik =W iK *p ik Among them, p ik It represents the final probability of the i-th test sample in the k-th classification category.
4. The dynamic classification method according to any one of claims 1 to 3, It is characterized in that Based on the final probability and the number of samples in each category after each sub-classifier classifies the training set, the decision threshold of each sub-classifier is determined, including: Perform fuzzy clustering on the disease data in the training set to obtain the number of samples N1 and N0 in two categories; For a subclassifier, calculate the ratio of the number of samples of the two categories I1 = N1 / N0, and the final probability ratio of the i-th test sample in the two categories in the classifier I2 = P i1 / P i0 ; Determine the size of the sample number ratio I1 and the final probability ratio I2; When the final probability ratio I2 is greater than the sample number ratio I1, the test sample is determined to be of category 1; otherwise, the test sample is determined to be of category 0.
5. The dynamic classification method according to claims 1-3, It is characterized in that After each sub-classifier performs final classification on the test sample according to the determined decision threshold, the step of evaluating the sub-classifier pool according to a pre-configured evaluation index is also included, and the evaluation index includes an index CE: Where N is the total number of samples of disease data in the test set; K is the number of categories, K = 2; a + Indicates that weight is given The sensitivity indicates the classification accuracy of class 1 samples; N1 indicates the number of samples classified into class 1 in the test set; a - Indicates that weight is given , which indicates the classification accuracy of class 0 samples; N0 indicates the number of samples classified as class 0 in the test set, TP indicates the number of samples correctly predicted as class 1, FN indicates the number of samples incorrectly predicted as class 0, TN indicates the number of samples correctly predicted as class 0, and TN indicates the number of samples incorrectly predicted as class 1.
6. A dynamic classification device for improving integration strategies for disease data, It is characterized in that include: A data acquisition module is used to obtain the disease data to be classified and split it into a training set and a test set according to a preset ratio; A fuzzy clustering module, used to perform fuzzy clustering on the disease data contained in the training set into two types of samples, and find a preset number of neighboring samples of the two cluster centers as verification samples; A sub-classifier pool forming module, used for inputting the verification samples into a pre-configured weak classifier set containing different types of classifiers, and selecting a preset number of classifiers from the weak classifier set according to the classification results to form a sub-classifier pool; A classification probability determination module is used to obtain the weight of a test sample in the test set on each category according to the cluster center, input the test sample into the sub-classifier pool, and obtain the predicted probability of the test sample on different categories; and to combine the weight and the predicted probability to obtain the final probability of the test sample on different categories; A decision threshold determination module, used to determine the decision threshold of each sub-classifier based on the final probability and the number of samples in each category after the training set is classified; The classification module is used to perform the final classification of the test samples according to the determined decision threshold.
7. The dynamic classification device according to claim 6, It is characterized in that The classification probability determination module includes a weight determination unit for obtaining weights W of different classifications in the subclassifier. ik : Among them, F represents the characteristic dimension of the test samples in the test set; x if represents the fth attribute of the i-th test sample; c kf represents the fth attribute of the kth class center, k=0 represents category 0, and k=1 represents category 1.
8. The dynamic classification device according to claim 7, It is characterized in that The classification probability determination module comprises: A weight matrix determination unit is used to determine the weights W of different categories in a subclassifier. ik Get the sample weight matrix W of the test sample iK : IN iK =[W i0 ,IN i1 ] Where K represents the number of categories, K = 2; W i0 represents the weight of the 0 category of the i-th test sample, W i1 represents the weight of category 1 of the i-th test sample; The final probability determination unit is used to obtain the sample weight matrix W iK And the predicted probability obtains the final probability P on the corresponding sub-classifier ik : P ik =W iK *p ik Among them, p ik It represents the predicted probability of the i-th test sample in the k-th classification category.
9. A dynamic classification device as claimed in any one of claims 6 to 8, It is characterized in that The decision threshold determination module comprises: The sample number determination unit is used to perform fuzzy clustering on the disease data in the training set to obtain the sample numbers N1 and N0 of two categories; A ratio determination unit is used to calculate the ratio of the number of samples of two categories for a sub-classifier. I1 = N1 / N0, and the final probability ratio of the i-th test sample in the two categories in the classifier is I2 = P i1 / P i0 ; The decision threshold judgment unit is used to judge the size of the sample number ratio I1 and the final probability ratio I2; when the final probability ratio I2 is greater than the sample number ratio I1, the test sample is judged to be category 1, otherwise the test sample is judged to be category 0.
10. A dynamic classification device as claimed in claims 6 to 8, It is characterized in that The dynamic classification device further includes an evaluation module for evaluating the sub-classifier pool according to pre-configured evaluation indicators, wherein the evaluation indicators include indicator CE: Where N is the total number of samples of disease data in the test set; K is the number of categories, K = 2; a + Indicates that weight is given The sensitivity indicates the classification accuracy of class 1 samples; N1 indicates the number of samples classified into class 1 in the test set; a - Indicates that weight is given , which indicates the classification accuracy of class 0 samples; N0 indicates the number of samples classified as class 0 in the test set, TP indicates the number of samples correctly predicted as class 1, FN indicates the number of samples incorrectly predicted as class 0, TN indicates the number of samples correctly predicted as class 0, and TN indicates the number of samples incorrectly predicted as class 1.
11. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor runs the computer program, the steps of the dynamic classification method according to any one of claims 1 to 5 are implemented.
12. A computer-readable storage medium storing a computer program. It is characterized in that When the computer program is executed by a processor, the steps of the dynamic classification method according to any one of claims 1 to 5 are implemented.
Citation Information
Patent Citations
Unbalanced data classification system based on geometric structure integration
CN109492096A
Graph-based dynamic classifier selection method
CN110866463A
Test data verification method based on ensemble learning and decision threshold change
CN111858568A
Selective integration method based on dynamic classifier sequence combination
CN112434734A
Breast cancer survival prediction method based on improved AdaBoost
CN114898869A
Cited By
Data cleaning method for clinical test EDC system
CN119274725A
Backdoor poisoning defense method and system for communication signal intelligent identification
CN119669762A
Method and device for training user classification model
CN121052871A
