CN115206538A - Perioperative patient sample data set balancing method and sample data set acquisition system - Google Patents

Abstract

The invention provides a method for balancing a patient sample data set in a perioperative period and a sample data set acquiring system. The sample data set equalization method includes: S1, over-sampling the minority label samples in the perioperative patient's sample data set to obtain synthetic samples, and generating a corresponding synthetic label set for the synthetic samples, and the sample data set includes a plurality of samples corresponding to the samples. Classification label set; S2, adding synthetic samples and synthetic label sets to the sample data set to obtain a temporary sample data set; S3, cleaning the samples in the temporary sample data set to obtain a balanced sample data set. Oversampling the minority class label samples in the sample data set to increase the number of minority class label samples, balance the majority class label samples and minority class label samples, clean the noise samples to improve the sample quality in the output balanced sample data set, and use the balanced sample data set for Subsequent classification processing can improve the performance of the classification model.

Description

A method for balancing the sample data set of perioperative patients and a sample data set acquisition system

technical field

The invention relates to the field of computer technology, in particular to a method for balancing a patient sample data set in a perioperative period and a sample data set acquisition system.

Background technique

The perioperative period is the perioperative period, and the perioperative period is a whole process surrounding the operation, starting from the patient's decision to receive surgical treatment, to the surgical treatment until the basic recovery, including the period before, during and after the operation. From the time when the surgical treatment is determined, until the treatment related to this surgery is basically finished, the time is about 5-7 days before surgery to 7-12 days after surgery.

According to the World Health Statistics 2021 report released by the World Health Organization (WHO), the life expectancy of the global population has increased to 73.3 years. The growing geriatric population around the world has been identified as the major population in the surgical market, and risk event prediction in geriatric patients has become one of the hot research directions. Postoperative risk prediction for elderly surgical patients can help doctors to formulate diagnosis and treatment plans, rationally allocate rescue resources, and reduce the probability of postoperative risk events. At present, some diagnostic tools can help hospitals provide comprehensive and reliable treatment for high-risk patients. For example, Chinese patents with publication numbers CN111009322A and CN114038565A have disclosed the use of predictive models for perioperative risk assessment based on patient perioperative data sets. However, , In the perioperative data set of patients, there is often a problem of imbalanced labels in the data set, which will directly affect the performance of the perioperative prediction model.

SUMMARY OF THE INVENTION

The invention aims to solve the technical problems existing in the prior art, and provides a method for balancing a patient sample data set in a perioperative period and a sample data set acquisition system.

In order to achieve the above object of the present invention, according to the first aspect of the present invention, the present invention provides a method for balancing patient sample data sets in the perioperative period, comprising: step S1, using the MLSMOTE algorithm to align the sample data sets of perioperative patients in the perioperative period. The minority class label samples are oversampled to obtain synthetic samples, and a corresponding synthetic label set is generated for the synthetic samples, and the sample data set includes a plurality of samples and the corresponding classification label sets of the samples; Step S2, the synthetic samples are added to the sample data set to obtain temporary Sample data set; Step S3, cleaning the samples in the temporary sample data set to obtain a balanced sample data set.

The above technical solution: oversampling the minority class label samples in the sample data set to increase the number of minority class label samples to achieve a balance between the majority class label samples and the minority class label samples. The generated noise samples are cleaned in all samples to improve the quality of the samples in the output balanced sample data set and effectively enhance the data. When the balanced sample data set is used for subsequent classification processing, the performance of the classification model can be improved.

In order to achieve the above object of the present invention, according to a second aspect of the present invention, the present invention provides a sample data set equalization device for perioperative patients, including: a sample synthesis module, which uses the MLSMOTE algorithm to analyze the samples of perioperative patients. The minority label samples in the data set are oversampled to obtain synthetic samples, and a corresponding synthetic label set is generated for the synthetic samples. The sample data set includes a plurality of samples and the corresponding classification label sets of the samples; the temporary sample data set acquisition module is used to synthesize the samples. Add the sample data set to obtain a temporary sample data set; the cleaning module cleans the samples in the temporary sample data set to obtain a balanced sample data set.

The above technical solution: the MLSMOTE algorithm is used to oversample the minority class label samples in the sample data set to increase the number of minority class label samples and achieve a balance between the majority class label samples and the minority class label samples. The noise samples generated in the generation process are cleaned in all samples, which improves the sample quality in the output balanced sample data set, effectively enhances the data, and can improve the performance of the classification model when the balanced sample data set is used for subsequent classification processing.

In order to achieve the above objects of the present invention, according to a third aspect of the present invention, the present invention provides a system for acquiring a perioperative patient sample data set, comprising: a data acquisition module for acquiring the original perioperative period of a plurality of patients Feature data and cases; a classification label set acquisition module, which obtains a classification label set based on multiple cases, and classification labels represent perioperative patient risk events; classification label association module, which is used to set the patient's original perioperative characteristic data and classification labels. At least one classification label is associated and corresponds; the perioperative patient data dimensionality reduction device performs dimensionality reduction processing on the original perioperative characteristic data of all patients to obtain the corresponding perioperative characteristic data; the sample data set acquisition module is based on the patient's perioperative characteristic data. The perioperative feature data is used as a sample, and the classification label set corresponding to the corresponding original perioperative feature data is associated with the sample to obtain the perioperative patient sample data set; and the perioperative patient sample according to the second aspect of the present invention is also included. The data set equalization device is used to perform equalization processing on the sample data set.

The above technical solution: a multi-class label sample data set of perioperative patients is constructed, and the perioperative patient data dimension reduction device makes the feature dimension of the samples in the data set lower and has a greater impact on the subsequent classification, which can speed up the process. Efficiency of subsequent classification processing and model training; increasing the number of minority-labeled samples through the sample data set equalization device to achieve a balance between majority-labeled samples and minority-labeled samples, and compared with MLSMOTE in the process of generating minority-labeled samples. Noise samples are cleaned in all samples to improve the sample quality in the output balanced sample data set, effectively enhancing the data. When the balanced sample data set is used for subsequent classification processing, the performance of the classification model can be improved.

Description of drawings

1 is a schematic structural diagram of a perioperative patient data dimension reduction device in Embodiment 1 of the present invention;

2 is a schematic structural diagram of a system for acquiring a patient sample data set in the perioperative period in Embodiment 2 of the present invention;

3 is a schematic flowchart of a method for equalizing a sample data set in Embodiment 3 of the present invention;

4 is a schematic structural diagram of a sample data set equalization device in Embodiment 4 of the present invention;

5 is a schematic structural diagram of a sample data set acquisition system in Embodiment 5 of the present invention;

6 is a schematic flowchart of a multi-label classification method for perioperative patient data in Embodiment 6 of the present invention;

Fig. 7 is the structural representation of classification model among the embodiment 6;

Fig. 8 is a kind of preferred flow chart of the multi-label classification method of perioperative patient data in embodiment 6;

9 is a schematic structural diagram of a multi-label classification device for perioperative patient data in Embodiment 7 of the present invention;

FIG. 10 is a schematic structural diagram of a perioperative patient risk event prediction system in Embodiment 8 of the present invention.

Detailed ways

The following describes in detail the embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary, only used to explain the present invention, and should not be construed as a limitation of the present invention.

In the description of the present invention, it should be understood that the terms "portrait", "horizontal", "upper", "lower", "front", "rear", "left", "right", "vertical", The orientations or positional relationships indicated by "horizontal", "top", "bottom", "inside", "outside", etc. are based on the orientations or positional relationships shown in the accompanying drawings, which are only for the convenience of describing the present invention and simplifying the description, rather than An indication or implication that the referred device or element must have a particular orientation, be constructed and operate in a particular orientation, is not to be construed as a limitation of the invention.

In the description of the present invention, unless otherwise specified and limited, it should be noted that the terms "installed", "connected" and "connected" should be understood in a broad sense, for example, it may be a mechanical connection or an electrical connection, or two The internal communication between the elements may be directly connected or indirectly connected through an intermediate medium. For those of ordinary skill in the art, the specific meanings of the above terms can be understood according to specific situations.

Example 1

This embodiment discloses a perioperative patient data dimension reduction device, as shown in FIG. 1 , the device includes:

The input module obtains the original perioperative characteristic data containing multi-dimensional features of the patient, and the classification label corresponding to the original perioperative characteristic data;

The initial dimension reduction module, based on the principal component analysis algorithm, performs dimension reduction processing on the original perioperative characteristic data to obtain the first perioperative characteristic data;

The second dimension reduction module, based on the genetic algorithm, performs dimension reduction processing on the characteristic data of the first perioperative period to obtain the characteristic data of the perioperative period;

The output module outputs perioperative characteristic data.

In this embodiment, in order to better reflect the patient's perioperative state, improve the accuracy of subsequent classification and processing, and avoid the problem of difficult collection and management of postoperative patient data, preferably, the original perioperative characteristic data includes the patient's preoperative, Intraoperative index data, such as preoperative blood pressure, heart rate, blood lipids, etc., intraoperative heart rate, blood pressure, blood loss, operation time, etc. Unlike some existing classification prediction models that only include the preoperative basic conditions of surgical patients and do not consider the specific conditions of the ongoing operation, many studies have confirmed that intraoperative indicators such as intraoperative heart rate, blood pressure, blood loss, and operation time are all consistent with each other. It is related to the postoperative condition of the patient, so the original perioperative characteristic data provided in this embodiment can improve the accuracy of the subsequent model in predicting postoperative events, and does not rely on postoperative patient index data.

In this embodiment, the classification label is used to characterize perioperative patient risk events, and perioperative patient risk events preferably include, but are not limited to, unplanned readmission and death.

In this embodiment, in order to improve the richness of the data, the index data includes category data and numerical data, and the category data is the index data represented by categories. Metric data, such as blood pressure values.

In this embodiment, the original perioperative characteristic data may be data of patients with known perioperative patient risk events. Therefore, the known perioperative patient risk events may be associated with the corresponding classification as the original perioperative characteristic data. Label. The original perioperative characteristic data may also be data of patients with unknown perioperative patient risk events, and experts set corresponding classification labels to the original perioperative characteristic data. There can be one, two or more classification labels corresponding to the original perioperative characteristic data.

In this embodiment, after processing by the principal component analysis algorithm, the feature dimension of the first perioperative feature data is smaller than the feature dimension of the original perioperative feature data, and an initial population of the genetic algorithm is constructed based on the first perioperative feature data.

In this embodiment, in order to further reduce the dimension of the first perioperative feature data through the genetic algorithm, preferably, the second dimension reduction module includes:

The initial population setting unit is used to set individuals based on the characteristic data of the first perioperative period. The number of genes of the individual is less than or equal to the total number of characteristics in the characteristic data of the first perioperative period, and multiple individuals form the initial population; the genes of the individual are the first perioperative period. For the characteristics in the characteristic data, the number of genes of each individual can be randomly set under the condition that the number of genes of the individual is less than or equal to the total number of characteristics in the characteristic data of the first perioperative period;

Evolutionary iterative unit, repeat the following process until reaching the termination condition, and output the individual with the largest fitness when the termination condition is reached: obtain the fitness of each individual in the current generation population; select some individuals from the current generation population based on the fitness of the individual As an individual of the next generation population; perform crossover and mutation operations on the individuals of the next generation.

In this embodiment, the termination condition is preferably, but not limited to, that the number of evolution iterations has reached the preset maximum number of evolution iterations, or the maximum value of the fitness of the individual in the evolutionary iteration is no longer increased, or the fitness of the individual in the evolutionary iteration The maximum increase is below the increase threshold. In each iteration, the fitness of the individuals in the current generation population is sorted from high to low, and some of the top-ranked individuals are selected as the individuals of the next generation population. The crossover operation is mainly to exchange the same locus of the paired parent, and after the exchange, the offspring is obtained, and the offspring is regarded as the individual of the next generation population.

In this embodiment, in order to make the perioperative feature data after dimensionality reduction have a better performance in the subsequent classification processing and improve the classification accuracy, preferably, the process of obtaining the fitness of an individual is: obtaining the original data of multiple patients. Perioperative feature data and corresponding classification labels, perform dimensionality reduction processing on multiple original perioperative feature data according to individual feature information to obtain multiple dimensionality reduction samples consistent with individual characteristics; divide multiple dimensionality reduction samples into reduced dimensionality samples. Dimension training set and dimensionality reduction test set; construct dimensionality reduction multilayer perceptron neural network; use dimensionality reduction training set to train the constructed dimensionality reduction multilayer perceptron neural network to obtain dimensionality reduction classification prediction model; use dimensionality reduction test set to conduct dimensionality reduction The classification prediction model is tested to obtain the accuracy of the model, and the accuracy is taken as the fitness of the individual.

Example 2

This embodiment discloses a perioperative patient sample data set acquisition system, as shown in FIG. 2 , the perioperative patient sample data set acquisition system includes: a data acquisition module for acquiring the original perioperative period of multiple patients Feature data and cases; case data is generally text data, including doctor's diagnosis, past medical history, postoperative follow-up records, etc.; the classification label set acquisition module, based on multiple cases, obtains the classification label set, and the classification label represents the risk events of perioperative patients; The classification label association module is used to associate the original perioperative characteristic data of the patient with at least one classification label in the classification label set, so the original perioperative characteristic data corresponds to a classification label set, and the classification label set includes at least one classification label; And the perioperative patient data dimensionality reduction device provided in Example 1, performs dimensionality reduction processing on the original perioperative characteristic data of all patients to obtain corresponding perioperative characteristic data; the sample data set acquisition module is based on the patient's perioperative period The characteristic data is used as a sample, and the classification label set corresponding to the corresponding original perioperative characteristic data is associated with the sample to obtain a sample data set of perioperative patients.

In this embodiment, preferably, the classification label set acquisition module specifically performs: performing word segmentation processing on patient cases to obtain at least one postoperative event result (postoperative event result is the perioperative patient risk event), Post-event results Use the trained CBOW model to perform similar word analogy to obtain multiple similar post-operative event result sets, match the similar post-operative event result sets with the event dictionary, and find the classification matching the similar post-operative event result sets from the event dictionary. label, a plurality of classification labels constitute a classification label set.

In this embodiment, the CBOW Multi-Word Context Model model of Word2Vec is trained on a large number of medical corpora, and the PKUSEG word segmentation tool (PKUSEG can segment words in multiple fields, including independent models in the medical field) is used to implement this implementation. The text information corresponding to the case set in this example is processed by word segmentation to obtain multiple postoperative event results. The event dictionary is preferably, but not limited to, the Chinese version of the ICD-11 event dictionary of the Uniform International Classification of Diseases published by the World Health Organization, and the event dictionary contains many classification labels. Whether the similar postoperative event result set matches the event dictionary is preferably, but not limited to, judgment by semantic similarity. If the semantic similarity between the two is greater than the preset similarity threshold, the two are considered to match, otherwise they do not match.

In this embodiment, preferably, in order to fill in the missing values in the data and improve the data quality, a missing filling device is also included, which is used to fill in the missing values in the original perioperative characteristic data of the patient, and The original perioperative characteristic data after filling and processing is input into the perioperative patient data dimensionality reduction device for dimensionality reduction processing. The missing filling device preferably, but is not limited to, performs filling processing through the existing RandomForestRegressor filling method, Missforest filling method, Mean filling method, or median filling method.

In this embodiment, further preferably, the missing filling device performs missing filling processing on the original perioperative feature data based on a Bayesian Gaussian process latent variable model.

In this embodiment, data imputation for missing values inevitably introduces uncertainty into the original perioperative feature dataset. This embodiment uses the Bayesian Gaussian process latent variable model (BGPLVM) to fill in the missing values of the numerical features, specifically including:

其中

是向量y_*中可以观察到的值，

是需要预测的缺失值。给定部分观察到的点y_*，本实施例希望重建丢失的部分

通过在一个小型完整数据集上学习对可观察变量的低维embedding，来填补缺失数据集。将BGPLVM在完整的数据集D上进行训练，引入隐变量X和新的测试隐变量x_*，如前所述

表示单个病人测量值的行向量，

代表已知观测值，

表示缺失值，通过最大化下面概率密度，得到y_*对应的隐变量x_*的高斯概率分布。First, approximately calculate the probability density p(y _* |Y) of the observed test data vector y _* ∈ R ^{N × M} (where N is the total number of patient samples and M is the total number of features), the implicit value associated with the observed value y _* The variational distribution of the variable is q(x _* ). After model parameters and latent variables are learned, BGPLVM can be used to estimate missing values:

in

is the observable value in the vector y _* ,

is the missing value that needs to be predicted. Given a partially observed point y _* , this embodiment wishes to reconstruct the missing part

Impute missing datasets by learning low-dimensional embeddings of observable variables on a small full dataset. The BGPLVM is trained on the complete dataset D, introducing a latent variable X and a new test latent variable x _* , as previously described

a row vector representing individual patient measurements,

stands for known observations,

Representing missing values, the Gaussian probability distribution of the latent variable x _* corresponding to y _* is obtained by maximizing the probability density below.

的变分下界来优化变分分布q(x_*)，保持除q(x_*)之外的所有优化量不变。为了预测缺失值

本发明采用标准高斯过程预测方法，同时将输入x_*的不确定因素也考虑进去，因为x_*存在分布q(x_*)。与GP预测形式相似，为了预测

本发明先预测

即与y_*对应的隐函数值

Next, by maximizing the

to optimize the variational distribution q(x _* ), keeping all optimization quantities except q(x _* ) constant. To predict missing values

The present invention adopts the standard Gaussian process prediction method, and simultaneously takes into account the uncertain factors of the input x _* , because x _* has a distribution q(x _* ). Similar to the GP prediction form, in order to predict

The present invention first predicts

i.e. the implicit function value corresponding to y _*

是可以分析处理的，在本发明中本发明用到了

的均值和协方差，均值可以为本发明提供缺失值的估计，方差则可以量化与均值估计相关的不确定性。通过BGPLVM模型，在训练集学习得到的隐空间和模型超参数，通过分布得到对于每个包含缺失值的特征的平均估计。Marginalization on x _* yields a multivariate density that is not fully Gaussian dependent, but based on a squared exponential kernel,

can be analyzed and processed, in the present invention, the present invention uses

The mean and covariance of the mean can provide the present invention with an estimate of missing values, and the variance can quantify the uncertainty associated with the mean estimate. Through the BGPLVM model, the latent space and model hyperparameters learned on the training set are distributed to obtain an average estimate for each feature containing missing values.

In this embodiment, in order to facilitate data processing, it is further preferable to further include an encoding device for encoding the original perioperative characteristic data, and inputting the encoded data into the missing filling device. The encoding device preferably, but is not limited to, uses the existing One-hot encoding rules for encoding.

In this embodiment, in order to facilitate data processing, further preferably, a normalization device is further included, which is used to normalize the encoded original perioperative characteristic data, and normalize the normalized data. Enter the missing padding device. The normalization device preferably, but not limited to, uses the standard deviation normalization method to perform normalization processing.

Example 3

This embodiment provides a method for balancing sample data sets of perioperative patients. As shown in FIG. 3 , the method for balancing sample data sets includes:

Step S1, performing oversampling on the minority label samples in the perioperative patient's sample data set to obtain synthetic samples, and generating a corresponding synthetic label set for the synthetic samples, the sample data set includes a plurality of samples and a classification label set corresponding to the samples; Each sample represents the perioperative feature data set of a patient, which can be the original perioperative feature data or the perioperative feature data obtained after dimension reduction of the original perioperative feature data in Example 1. The classification and label association process of the samples It has been described in detail in Embodiment 1 and will not be repeated here.

Step S2, adding the synthetic sample and the synthetic label set to the sample data set to obtain a temporary sample data set;

Step S3, cleaning the samples in the temporary sample data set to obtain a balanced sample data set.

In this embodiment, synthetic samples can be obtained by over-sampling the minority-labeled samples in the perioperative patient sample data set through SMOTE or SVM SMOTE or BorderlineSMOTE or K-Means SMOTE or SMOTE-NC, and the corresponding synthetic samples can be generated for the synthetic samples. Synthetic label set. Preferably, in order to improve the balance effect, the MLSMOTE algorithm is used to oversample the minority label samples in the perioperative patient's sample data set to obtain synthetic samples, and to generate corresponding synthetic label sets for the synthetic samples. The MLSMOTE algorithm is the Multi-label Synthetic Minority Over-sampling Technique (MLSMOTE), which is often used to deal with the problem of data imbalance in multi-label classification tasks. The generation process includes: using the Imbalance Rate (IR) Select minority class labels; Nearest neighbor search: Once a sample belonging to a minority label is selected as a seed sample, its nearest neighbors are searched; Feature set generation: After selecting a neighborhood, synthetic samples are obtained by interpolation; Synthetic label set generation : A synthetic label set is required for the generated synthetic samples.

In this embodiment, since the oversampling synthesizing minority class sample algorithm such as MLSMOTE will generate some noise samples in the process of synthesizing minority class label samples, it is necessary to clean these noise samples, so step S3 is set to improve the quality of the sample data set .

In this embodiment, preferably, in order to quickly determine the minority class labels in the sample data set, the ratio of the number of samples corresponding to each class label to the total number of samples in the sample data set is calculated, and the class label whose ratio is less than the ratio threshold is used as Minority class classification label, the classification label greater than or equal to the ratio threshold is used as the majority class classification label, and the ratio threshold is preferably but not limited to less than 0.2.

In this embodiment, the number of samples to be generated for each minority class classification label is the oversampling rate of the minority class classification label. In order to better determine the oversampling rate of each minority class classification label, so that the obtained balanced sample data set performs better when applied to subsequent classification, preferably, in step S1, based on the genetic algorithm, each minority class label is assigned. Set the oversampling rate, including:

Step S11, set the sample data set to include W minority class labels, take the oversampling rate of the samples with W minority class labels as individual W genes, and W is a positive integer; each gene represents an oversampling of a minority class label. rate, using multiple individuals to construct an initial population, the initial population includes multiple initial individuals, the value of W genes of each initial individual is obtained by random selection, preferably, it can be set for the oversampling rate of each minority class classification label Numerical range, when constructing the initial population, the numerical value is randomly selected as the gene value in this numerical range, and the numerical range can be set as required;

In step S12, the following evolutionary iterative process is repeatedly performed until the termination condition is reached: obtaining the fitness of each individual in the current generation population; selecting some individuals from the current generation population based on the fitness of the individuals as individuals of the next generation population; Individuals of , perform crossover and mutation operations;

Step S13, output the individual with the largest fitness when the termination condition is reached.

In this embodiment, the termination condition is preferably, but not limited to, that the number of evolution iterations has reached the preset maximum number of evolution iterations, or the maximum fitness of the individual in the evolutionary iteration is no longer increased, or the fitness of the individual in the evolutionary iteration is the maximum The value increases by less than the increase threshold. In each iteration, the fitness of the individuals in the current generation population is sorted from high to low, and some of the top-ranked individuals are selected as the individuals of the next generation population.

In this embodiment, in order to make the obtained balanced sample data set perform better when applied to subsequent classification, preferably, the process of obtaining the fitness of an individual:

Obtain the minority class label oversampling rate combination based on individual genetic information; the oversampling rate combination includes the oversampling rate of all minority class labels;

Based on the combination of minority class label oversampling rate, the minority class label samples in the perioperative patient's sample data set are oversampled to obtain synthetic samples and synthetic label sets of synthetic samples, and the synthetic samples and synthetic label sets are added to the sample data set to obtain balanced samples The balanced sample set is divided into a balanced training sample set and a balanced test sample set;

Construct a balanced multi-layer perceptual neural network, use the balanced training sample set to train the balanced multi-layer perceptual neural network to obtain the balanced prediction classification model, use the balanced test sample set to test the balanced prediction classification model to obtain the accuracy of the balanced prediction classification model, and use the accuracy as individual fitness.

In this embodiment, in order to effectively remove noise samples and improve the quality of the sample set, preferably, step S3 is to perform cleaning processing on each sample in the temporary sample data set, and the cleaning processing process includes:

Step S31, select a seed sample from the temporary sample data set, select k neighbor samples of the seed sample, and the classification labels of the k neighbor samples form a neighbor classification label set, and k is a positive integer; each sample in the temporary sample data set can be selected in turn. as a seed sample;

Step S32, predicting the classification label set of the seed sample by Bayesian conditional probability based on the nearest neighbor classification label set, and obtaining the predicted classification label set of the seed sample;

Step S33, determine whether the predicted classification label set of the seed sample is the same as the classification label set in the temporary sample data set, if the same, keep the seed sample, if not, delete the seed sample, and consider the seed sample to be a noise sample.

The above cleaning process directly predicts the classification label set of the seed sample through Bayesian conditional probability based on the nearest neighbor classification label set of the seed sample, and compares the obtained predicted classification label set with the real classification label set of the seed sample in the temporary sample data set. It will not rely on the classifier to determine, but only rely on the data itself to determine, reduce the amount of calculation, and improve the efficiency and accuracy of the judgment.

In this embodiment, further preferably, in step S31, the specific process of selecting the k nearest neighbor samples of the seed sample includes:

Obtain the heterogeneous value difference metric HVDM between the seed sample and all or part of the samples in the temporary sample data set; HVDM is the abbreviation of Heterogeneous Value Difference Metric;

Using the global imbalance weight of the samples in the temporary sample data set to modify the outlier difference measure HVDM to obtain the revised outlier difference measure;

Sort the modified outlier difference metric of all samples and seed samples in the temporary sample data set, and select the first k samples with larger modified outlier difference metric as the k nearest neighbor samples of the seed sample. Preferably, the modified outlier difference metric may be sorted from high to bottom, and the first k samples with larger modified outlier difference metric values are selected as the k nearest neighbor samples of the seed sample.

The weighted KNN (Weighted kNN, WkNN) method is used to improve the quality of the synthesized samples in the process of selecting the k nearest neighbor samples of the seed sample. If the distribution of the real minority class label samples in the sample data set is very scattered, that is, the space is sparse, the minority class samples synthesized during the execution of algorithms such as MLSMOTE will still be scattered and sparse, and there is still no balance from a local perspective. If kNN is used for cleaning directly, the sparse minority samples and the new minority samples synthesized by MLSMOTE will be removed with a high probability, so that an appropriate classification boundary cannot be established. Therefore, the idea of distance weighting needs to be introduced to coordinate kNN cleaning, that is, When faced with sparsely distributed samples, it is not blindly deleted directly, but the local spatial density (that is, the heterogeneous value difference measurement HVDM and the global imbalance weight of the sample) is taken into account, and the small samples are retained as much as possible. The cleaning of kNN mainly relies on the label set of the neighbor samples, so the distance calculation for the neighbors is particularly important when the data distribution is sparse. the main reason for the correction). WkNN is used to clean the noise samples and change the distance of calculating the nearest neighbor samples (correction of the modified outlier difference metric), that is, considering the influence of local density, the sample anomaly value difference metric is used to represent the distance between samples.

In this embodiment, further preferably, the calculation formula of the heterogeneous value difference measurement HVDM of the samples in the seed sample and the temporary sample data set is:

C表示当特征x为类别特征时该特征的类别数，c表示特征x的类别索引，

表示临时样本数据集中特征x属于特征向量f₁且特征x的类别特征为c的样本数；

表示临时样本数据集中特征x属于特征向量f₂且特征x的类别特征为c的样本数；

表示临时样本数据集中特征x属于特征向量f₁的样本数；

表示临时样本数据集中特征x属于特征向量f₂的样本数；|f₁-f₂|表示特征向量f₁与f₂差值的绝对值；σ_x表示临时样本数据集中特征x的标准差。Among them, f ₁ represents the feature vector of the seed sample; f ₂ represents the feature vector of any sample in the temporary sample data set except the seed sample; HVDM(f ₁ , f ₂ ) represents the heterogeneous value of the feature vector f ₁ and f ₂ Difference metric; D(f ₁ , f ₂ ) represents the distance between the feature vectors f ₁ and f ₂ ; n represents the feature dimension of the sample in the temporary sample dataset; x represents the feature index; d _x (f ₁ , f ₂ ) Represents the distance between the feature vector f ₁ and the feature vector f ₂ on the feature x, d _x (f ₁ , f ₂ ) is obtained by the following formula:

C represents the number of categories of the feature when the feature x is a category feature, c represents the category index of the feature x,

Indicates the number of samples in the temporary sample dataset where feature x belongs to feature vector f ₁ and the category feature of feature x is c;

Indicates the number of samples in the temporary sample dataset where the feature x belongs to the feature vector f ₂ and the category feature of the feature x is c;

Indicates the number of samples whose feature x belongs to feature vector f ₁ in the temporary sample data set;

Represents the number of samples in the temporary sample data set where the feature x belongs to the feature vector f ₂ ; |f ₁ -f ₂ | represents the absolute value of the difference between the feature vector f ₁ and f ₂ ; σ _x represents the standard deviation of the feature x in the temporary sample data set.

In this embodiment, further preferably, the calculation formula of the modified outlier difference metric of the samples in the seed sample and the temporary sample data set is:

Among them, f ₁ represents the feature vector of the seed sample; f ₂ represents the feature vector of any sample in the temporary sample data set except the seed sample; HVDM(f ₁ , f ₂ ) represents the heterogeneous value of the feature vector f ₁ and f ₂ Difference metric; D _W (f ₁ , f ₂ ) represents the modified outlier difference metric of the feature vector f ₁ and f ₂ ; n represents the feature dimension of the sample in the temporary sample data set; IW represents the global value of the sample whose feature vector is f ₂ Imbalance weight, IW=IR _nn /(IR ⁺ +IR ^- ), IR ⁺ represents the total imbalance rate of all minority class labels in the temporary sample data set, IR ^- represents the total imbalance rate of all majority class labels in the temporary sample data set , IR _nn is the total imbalance rate of all classification labels in the classification label set of the sample whose feature vector is _f2 .

In the above process of removing noise samples, WkNN uses Heterogeneous ValueDifference Metric (HVDM) to measure the distance when calculating the distance, and uses the global imbalance weight IW of the sample as the weight coefficient to correct the HVDM. For the temporary sample data set, when the classification label set contains more minority class labels, the larger the IR _nn will be, the larger the IW will be; for the temporary sample data set with sparse minority class label sample distribution and large imbalance rate, the IW is introduced into the HVDM distance to increase the minority class sample density.

可以看到，加权系数

的值可以缩放HVDM(f₁,f₂)，近邻样本分类标签集中的少数类标签越多则加权系数

会越小。在种子样本的近邻样本集的IW越大时，即近邻样本标签集包含的少数类标签越多时，对应的近邻样本的加权系数

就越小，这样呈现单调递减的形式，可以维持：在特征维度固定的情况下，近邻样本的加权系数

会因为其标签集中包含的多数类和少数类标签的情况而不同程度的放缩；当特征维度增多时，也就是样本分布逐渐稀疏时，放缩系数也跟着变小。from the formula

It can be seen that the weighting factor

The value of can scale HVDM(f ₁ ,f ₂ ), the more the minority class labels in the nearest neighbor sample class label set, the weighting coefficient

will be smaller. When the IW of the neighbor sample set of the seed sample is larger, that is, when the label set of the neighbor sample contains more minority class labels, the weighting coefficient of the corresponding neighbor sample

The smaller it is, it presents a monotonically decreasing form, which can be maintained: when the feature dimension is fixed, the weighting coefficient of the nearest neighbor samples

It will be scaled to varying degrees because of the majority and minority labels contained in its label set; when the feature dimension increases, that is, when the sample distribution becomes sparse, the scaling factor also decreases.

It can be seen that WkNN can help to filter the neighbor samples for the samples with more minority class labels in the label set, taking into account the distribution of labels in the label set of the neighbor samples, so that the samples with more minority class labels in the label set move closer to the seed samples and increase Local minority class label density while reducing majority class label density. The overall process is as follows: First, use MLSMOTE to upsample the samples of minority class labels, and form a more balanced temporary new sample set with the original samples. On this new sample set, perform the WkNN process for each sample, that is, the weighted HVDM Sort out k nearest neighbor samples, and then predict the label set of the seed sample according to the nearest neighbor samples. If the predicted label set is the same as the seed label set, keep the sample, otherwise delete it

Example 4

This embodiment discloses a sample data set equalization device for perioperative patients. As shown in FIG. 4 , the sample data set equalization device includes:

The sample synthesis module is used to oversample the minority label samples in the perioperative patient's sample data set to obtain synthetic samples, and generate corresponding synthetic label sets for the synthetic samples. The sample data set includes multiple samples and the corresponding classification label sets of the samples;

The temporary sample data set acquisition module adds synthetic samples and synthetic label sets to the sample data set to obtain a temporary sample data set;

The cleaning module cleans the samples in the temporary sample data set to obtain a balanced sample data set.

In this embodiment, preferably, the cleaning module includes:

The nearest neighbor sample acquisition unit selects a seed sample from the temporary sample data set, selects k nearest neighbor samples of the seed sample, and the classification labels of the k nearest neighbor samples form a nearest neighbor classification label set, where k is a positive integer;

The predicted classification label set obtaining unit, based on the nearest neighbor classification label set, predicts the classification label set of the seed sample through Bayesian conditional probability, and obtains the predicted classification label set of the seed sample;

The cleaning unit determines whether the predicted classification label set of the seed sample is the same as the classification label set in the temporary sample data set, if the same, the seed sample is retained, if not, the seed sample is deleted.

In this embodiment, further preferably, the specific process of selecting the k nearest neighbor samples of the seed sample by the nearest neighbor sample acquisition unit includes:

Obtain the heterogeneous value difference metric HVDM between the seed sample and all or part of the samples in the temporary sample data set;

Using the global imbalance weight of the samples in the temporary sample data set to modify the outlier difference measure HVDM to obtain the revised outlier difference measure;

Sort the modified outlier difference metric of all samples and seed samples in the temporary sample data set, and select the first k samples with larger modified outlier difference metric as the k nearest neighbor samples of the seed sample.

The balancing effect of the sample data set balancing device provided in this embodiment is experimentally verified, and the results are as follows:

IR represents the Imbalance Rate of the sample set. The larger the IR, the more unbalanced the sample set is. From the experimental results in the table above, it can be seen that the maximum IR and average IR of the equalization device provided in this embodiment are the smallest, and the IR The interval between the maximum value and the mean value is narrowed, indicating that the sample set is better balanced.

Example 5

This embodiment also discloses a system for obtaining a sample data set of patients during the perioperative period. Compared with Embodiment 2, this embodiment adds a sample data set equalization device, that is, the sample data set obtained after dimensionality reduction obtained in Embodiment 2 is processed. For sample balance processing, the schematic diagram of the structure of the device is shown in Figure 5, including: a data acquisition module for acquiring the original perioperative characteristic data and cases of multiple patients; a classification label set acquisition module for obtaining classification labels based on multiple cases set, the classification labels represent perioperative patient risk events; the classification label association module is used to associate the original perioperative characteristic data of patients with at least one classification label in the classification label set; the perioperative patient data dimensionality reduction device is used for all The original perioperative characteristic data of the patient is subjected to dimensionality reduction processing to obtain the corresponding perioperative characteristic data; the sample data set acquisition module takes the patient's perioperative characteristic data as a sample, and associates the corresponding original perioperative characteristic data for the sample. The classification label set of the perioperative period is obtained to obtain the sample data set of the perioperative patient; and the device for balancing the sample data set of the perioperative patient provided in Embodiment 4 is also used for balancing the sample data set.

In this embodiment, preferably, a missing filling device is also included, which is used to fill in the missing values in the original perioperative characteristic data of the patient, and input the filled original perioperative characteristic data into the perioperative period. The patient data dimensionality reduction device performs dimensionality reduction processing.

Example 6

This embodiment 6 discloses a multi-label classification method for perioperative patient data, as shown in FIG. 6 , the multi-label classification method includes:

In step A, the characteristic data of the patient to be classified is obtained; the characteristic data of the patient to be classified is the characteristic data of the perioperative patient, which may include multi-dimensional features. In order to improve the processability, reduce the dimension, and improve the quality of the characteristic data of the patients to be classified, the characteristic data of the patients to be classified can be encoded, normalized, and output according to the perioperative patient data dimension reduction device provided in Example 1. The feature dimension of the sample is subjected to dimensionality reduction processing, and the feature data of the patient to be classified after the dimensionality reduction processing is input into the trained classification model.

Step B, input the characteristic data of the patient to be classified into the trained classification model, and the classification model outputs the classification result, and the classification result includes more than one classification label and the classification confidence of each classification label; the classification confidence of the classification label represents the characteristics of the patient to be classified. The probability that the data belongs to this classification label. The classification model includes a stacking-based classification integration model, a label association rule acquisition module, and a fusion module. The fusion module is used to fuse the classification matrix output by the classification integration model and the association rule matrix output by the label association rule acquisition module to obtain the classification results. The fusion method Preferably, but not limited to, the classification matrix and the association rule matrix are multiplied.

In the embodiment, preferably, a schematic diagram of the structure of the classification model is shown in FIG. 7 , and the classification integration model includes a first multi-classification model, a second multi-classification model, a third multi-classification model and a logistic regression model; the first multi-classification model , the second multi-classification model, and the third multi-classification model respectively perform multi-label classification processing on the patient characteristic data to be classified to obtain the first primary classification result, the second primary classification result, and the third primary classification result; the logistic regression model is used to classify the first primary classification The result, the second primary classification result, and the third primary classification result are processed to obtain a classification matrix.

In this embodiment, preferably, the first multi-classification model, the second multi-classification model, and the third multi-classification model are a Ranking-SVM model, a classification multilayer perceptual neural network model, and a Binary Relevance model, respectively. The Ranking-SVM model and the Binary Relevance model are the more conventional basic models in Stacking integration, and they are used here for high reliability of model integration. The classification multi-layer perceptual neural network model adopts the multi-layer perceptual neural network structure (ie, the MLP network structure), which can avoid the problem of overfitting and has low complexity.

In this embodiment, preferably, the step of constructing a sample data set of perioperative patients is also included. As shown in FIG. 8 , the step of constructing a sample data set of perioperative patients is preferably, but not limited to, using Embodiment 2 or implementing The system of Example 5 was constructed.

In the embodiment, as shown in FIG. 7 , the training process of the classification integrated model is: constructing a sample data set of perioperative patients, each sample in the sample data set is associated with more than one classification label, and the sample data set is divided into classification training sets and the classification test set, the association of classification labels can be performed manually; build a classification integration model, that is, the above-mentioned Stacking-based integration model, which includes the first multi-classification model, the second multi-classification model, the third multi-classification model and logistic regression Model; use the classification training set to train the classification ensemble model, and use the classification test set to test and verify the trained classification ensemble model. In validation, RandomizedSearchCV and GridSearchCV are used for cross-validation on the training set, and hyperparameter selection is performed by the F1_Micro score.

In this embodiment, as shown in FIG. 7 , preferably, the association rule acquisition module performs the following steps: acquiring a sample data set of patients in the perioperative period, each sample in the sample data set is associated with more than one classification label; the sample data set is preferably but It is not limited to the perioperative patient sample data set obtained in Example 2 or Example 5, that is, the standard patient data set. Perform association rule mining on the classification labels in the sample data set to obtain the association rule matrix. The association rule matrix includes the association confidence between any two class labels among all class labels.

In this embodiment, as shown in FIG. 7 , further preferably, when the number of classification labels in the sample data set is small, specifically when it is less than the number threshold, the classification labels in the sample data set are directly processed by the FP-growth algorithm. Association rule mining. First, establish a classification label matrix as shown in Figure 7, the first row of the classification label matrix is each label, and the first column is the patient number; after that, use the FP-growth algorithm to perform association rule analysis processing on the classification label matrix, and output any two The association confidence between the classification labels. The value of the association confidence ranges from 0 to 1. Based on these association confidences, an association rule matrix as shown in Figure 7 is established. In the association rule matrix, the first row and the first column are both classification labels, and the elements in the matrix represent the association between the classification labels of the row and column where the element is located. Confidence, as shown in Figure 7, A(N-1) represents the confidence of the association between the classification label N and the classification label 1.

In this embodiment, preferably, when the number of classification labels in the sample data set is large, the correlation patterns between the classification labels will be different, and the direct correlation analysis will cause the frequent itemset search process to be complicated, which affects the accuracy of the correlation analysis. , specifically, when the number of classification labels is greater than or equal to the number threshold, the number threshold is preferably, but not limited to, 3 or 4 or 5. The steps of performing association rule mining on the classification labels in the sample data set to obtain an association rule matrix include:

Clustering the classification labels in the sample data set to obtain more than one cluster; preferably but not limited to using the K-means++ algorithm for clustering processing; perform association rule mining on the classification labels in each cluster to obtain an association rule sub-matrix . During fusion, the classification matrix is divided into more than one sub-classification matrix according to the clustering results, one cluster cluster corresponds to one sub-classification matrix, and the sub-classification matrix is multiplied by the association rule sub-matrix corresponding to the cluster cluster to obtain the cluster The classification sub-results of the cluster, all the classification sub-results make up the classification result.

In this embodiment, it is further preferred to perform association rule mining on the classification labels in each classification cluster through the FP-growth algorithm to obtain an association rule sub-matrix. description, which is not repeated here.

Example 7

This embodiment discloses a multi-label classification device for perioperative patient data. As shown in FIG. 9 , it includes: a data acquisition module for acquiring characteristic data of patients to be classified; a classification module for inputting characteristic data of patients to be classified The trained classification model, the classification model outputs the classification result, and the classification result includes more than one classification label and the classification confidence of each classification label; the classification model includes the Stacking-based classification integration model, the label association rule acquisition module and the fusion module. The fusion module It is used to fuse the classification matrix output by the classification ensemble model and the label association rule acquisition module output association rule matrix to obtain the classification result.

In this embodiment, preferably, the classification integration model includes a first multi-class model, a second multi-class model, a third multi-class model and a logistic regression model; the first multi-class model, the second multi-class model, the third multi-class model The classification model separately performs multi-label classification processing on the patient characteristic data to be classified to obtain a first primary classification result, a second primary classification result, and a third primary classification result; the logistic regression model performs a multi-label classification process on the first primary classification result and the second primary classification result. , and the third primary classification result is processed to obtain a classification matrix.

In this embodiment, preferably, a classification integrated model training module is further included, and the classification integrated model training module performs the following process: constructing a sample data set of perioperative patients, each sample in the sample data set is associated with more than one classification label, and the sample The data set is divided into a classification training set and a classification test set; preferably but not limited to constructing a sample data set of perioperative patients through the system provided in Embodiment 2 or Embodiment 5; building a classification integrated model; the classification integrated model includes the first multi-classification model, the second multi-classification model, the third multi-classification model and the logistic regression model; use the classification training set to train the classification ensemble model, and use the classification test set to test and verify the trained classification ensemble model.

In this embodiment, the classification device builds a multi-label classification integration model of perioperative and postoperative events combined with association rule analysis. There may be a variety of postoperative risk events after surgery, and the results of multiple events after surgery are researched and predicted. By integrating the Ranking-SVM model, the multi-layer perceptual neural network model and the Binary Relevance model, a multi-label prediction model is built to further improve the model's performance. Stability and accuracy, it integrates association rules into the prediction model for optimization.

Example 8

This embodiment discloses a perioperative patient risk event prediction system, as shown in FIG. 10 , including: a data acquisition module for acquiring characteristic data of patients to be classified; a classification module for inputting characteristic data of patients to be classified into training A good classification model, the classification model outputs classification results, the classification results include more than one classification label and the classification confidence of each classification label, and each classification label corresponds to a perioperative patient risk event;

The classification model includes a stacking-based classification integration model, a label association rule acquisition module and a fusion module. The fusion module is used to fuse the classification matrix output by the classification integration model and the association rule matrix output by the label association rule acquisition module to obtain the classification result; the conversion module, Convert the classification labels in the classification results to the corresponding perioperative patient risk events to obtain risk prediction results.

In this embodiment, preferably, the classification integration model includes a first multi-class model, a second multi-class model, a third multi-class model and a logistic regression model; the first multi-class model, the second multi-class model, the third multi-class model The classification model separately performs multi-label classification processing on the patient characteristic data to be classified to obtain a first primary classification result, a second primary classification result, and a third primary classification result; the logistic regression model performs a multi-label classification process on the first primary classification result and the second primary classification result. , and the third primary classification result is processed to obtain a classification matrix.

In this embodiment, preferably, a classification integrated model training module is further included, and the classification integrated model training module performs the following process: constructing a sample data set of perioperative patients, each sample in the sample data set is associated with more than one classification label, and the sample The data set is divided into a classification training set and a classification test set; preferably but not limited to constructing a sample data set of perioperative patients through the system provided in Embodiment 2 or Embodiment 5; building a classification integrated model; the classification integrated model includes the first multi-classification model, the second multi-classification model, the third multi-classification model and the logistic regression model; use the classification training set to train the classification ensemble model, and use the classification test set to test and verify the trained classification ensemble model.

In this embodiment, in the process of obtaining the system sample data set provided in Embodiment 2 or Embodiment 5, the risk events in the perioperative period of patients (especially elderly surgical patients) are predicted, and the improvement of missing and unbalanced data sets is improved. Based on the analysis of association rules, a multi-label prediction model of postoperative events was built. Post-operative event label extraction is performed based on patient case text. The CBOW label extraction model of Word2Vec is used to collect a large number of medical-related corpora, train the medical word vector model, and realize the extraction of postoperative event label set (ie, classification label set). Next, the latent variable model based on Bayesian Gaussian process is used to fill missing data, and based on MLSMOTE, weighted kNN (WKNN) and genetic algorithm for label imbalance data processing, and finally combined with principal component analysis PCA model and genetic algorithm to build feature reduction dimensional model, providing more relevant input to the classification ensemble model.

Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, The scope of the invention is defined by the claims and their equivalents.

Claims (10)

1. A perioperative patient sample dataset equalization method, comprising:

the method comprises the following steps of S1, oversampling a few types of label samples in a sample data set of a perioperative patient to obtain a synthetic sample, and generating a corresponding synthetic label set for the synthetic sample, wherein the sample data set comprises a plurality of samples and a classification label set corresponding to the samples;

s2, adding the synthetic sample and the synthetic label set into the sample data set to obtain a temporary sample data set;

and S3, cleaning the samples in the temporary sample data set to obtain a balanced sample data set.

2. The perioperative patient sample dataset equalization method of claim 1, wherein in said step S1, setting an oversampling rate for each minority class label based on a genetic algorithm, specifically comprises:

s11, setting a sample data set to comprise W minority class labels, and taking the oversampling rate of the samples of the W minority class labels as W genes of an individual, wherein W is a positive integer; constructing an initial population, wherein the initial population comprises a plurality of initial individuals, and the W gene values of each initial individual are obtained by random selection;

step S12, the following evolutionary iterative process is repeatedly performed until a termination condition is reached:

acquiring the fitness of each individual in the population of the current generation; selecting a part of individuals from the population of the current generation as individuals of the population of the next generation based on the fitness of the individuals; performing cross operation and variation operation on individuals of the next generation population;

and S13, outputting the individual with the maximum fitness when the termination condition is reached.

3. The perioperative patient sample data set balancing method of claim 2, wherein the process of obtaining the fitness of an individual:

obtaining minority label oversampling rate combinations based on individual gene information; the over-sampling rate combination includes the over-sampling rates of all the minority class tags;

oversampling a few types of label samples in a sample data set of a perioperative patient based on a few types of label oversampling rate combination to obtain a synthetic sample and a synthetic label set of the synthetic sample, adding the synthetic sample and the synthetic label set into the sample data set to obtain an equilibrium sample set, and dividing the equilibrium sample set into an equilibrium training sample set and an equilibrium testing sample set;

the method comprises the steps of constructing an equilibrium multi-layer perception neural network, training the equilibrium multi-layer perception neural network by using an equilibrium training sample set to obtain an equilibrium prediction classification model, testing the equilibrium prediction classification model by using an equilibrium test sample set to obtain the accuracy of the equilibrium prediction classification model, and taking the accuracy as the fitness of an individual.

4. The perioperative patient sample dataset balancing method according to claim 1, 2 or 3, wherein the step S3 is a cleaning process for each sample in the temporary sample dataset, the cleaning process comprising:

s31, selecting seed samples from the temporary sample data set, selecting k adjacent samples of the seed samples, wherein classification labels of the k adjacent samples form an adjacent classification label set, and k is a positive integer;

step S32, predicting the classification tag set of the seed sample through Bayes conditional probability based on the neighbor classification tag set to obtain a predicted classification tag set of the seed sample;

and step S33, judging whether the predicted classification label set of the seed sample is the same as the classification label set of the seed sample in the temporary sample data set, if so, retaining the seed sample, and if not, deleting the seed sample.

5. The perioperative patient sample dataset equalization method of claim 4, wherein in said step S31, the specific process of selecting k neighbor samples of seed samples comprises:

obtaining heterogeneous value difference measurement HVDM of the seed sample and all or part of the samples in the temporary sample data set respectively;

correcting the heterogeneous value difference measurement HVDM by using the global unbalanced weight of the samples in the temporary sample data set to obtain a corrected heterogeneous value difference measurement;

and sequencing the correction heterogeneous value difference measures of all the samples and the seed samples in the temporary sample data set, and selecting the first k samples with larger correction heterogeneous value difference measures as k adjacent samples of the seed samples.

6. The perioperative patient sample dataset equalization method of claim 5, wherein the heterology difference measure HVDM of the seed sample and the temporary sample dataset is calculated by the formula:

wherein f is ₁ A feature vector representing a seed sample; f. of ₂ A feature vector representing any sample in the temporary sample data set except the seed sample; HVDM (f) ₁ ,f ₂ ) Representing a feature vector f ₁ And f ₂ A heterology difference metric of; d (f) ₁ ,f ₂ ) Representing a feature vector f ₁ And f ₂ The distance between them; n represents the characteristic dimension of the sample in the temporary sample data set; x represents a feature index; d _x (f ₁ ,f ₂ ) Representing a feature vector f ₁ And a feature vector f ₂ Distance, d, over feature x _x (f ₁ ,f ₂ ) Obtained by the following formula:

c denotes the number of classes of the feature x when the feature x is a class feature, C denotes a class index of the feature x,

representing the feature x in the temporary sample dataset as belonging to the feature vector f ₁ And the class feature of the feature x is the number of samples of c;

representing the feature x in the temporary sample dataset as belonging to the feature vector f ₂ And the class feature of the feature x is the number of samples of c;

representing the feature x in the temporary sample dataset as belonging to the feature vector f ₁ The number of samples of (a);

representing the feature x in the temporary sample dataset as belonging to the feature vector f ₂ The number of samples of (a); l f ₁ -f ₂ I represents a feature vector f ₁ And f ₂ The absolute value of the difference; sigma _x Representing the standard deviation of the feature x in the temporary sample dataset.

7. The perioperative patient sample dataset equalization method of claim 5 or 6, wherein the modified heterogeneous difference measure of the seed sample and the sample in the temporary sample dataset is calculated by the formula:

wherein f is ₁ A feature vector representing a seed sample; f. of ₂ A feature vector representing any sample in the temporary sample data set except the seed sample; HVDM (f) ₁ ,f ₂ ) Representing a feature vector f ₁ And f ₂ A heterogeneous value difference metric of (a); d _W (f ₁ ,f ₂ ) Representing a feature vector f ₁ And f ₂ Modified heterology difference metric of (a); n represents the characteristic dimension of the sample in the temporary sample data set; IW represents a feature vector of f ₂ Of samples of (a), IW = IR _nn /(IR ⁺ +IR ^- ),IR ⁺ Indicating the Total imbalance Rate, IR, of all the minority class Classification tags in the temporary sample dataset ^- Indicating the Total imbalance Rate, IR, of all of the majority class Classification tags in the temporary sample dataset _nn Is a feature vector of f ₂ The total imbalance rate of all the class labels in the class label set of the sample.

8. A perioperative patient's sample data set equalization apparatus, comprising:

the sample synthesis module is used for oversampling a few types of label samples in a sample data set of a perioperative patient to obtain a synthesized sample and generating a corresponding synthesized label set for the synthesized sample, wherein the sample data set comprises a plurality of samples and a classification label set corresponding to the samples;

the temporary sample data set acquisition module is used for adding the synthetic sample and the synthetic label set into the sample data set to acquire a temporary sample data set;

and the cleaning module is used for cleaning the samples in the temporary sample data set to obtain a balanced sample data set.

9. A perioperative patient sample dataset acquisition system, comprising:

the data acquisition module is used for acquiring original perioperative characteristic data and cases of a plurality of patients;

the classification label set acquisition module is used for acquiring a classification label set based on a plurality of cases, and the classification labels represent perioperative patient risk events;

the classification label association module is used for associating and corresponding original perioperative characteristic data of the patient with at least one classification label in the classification label set;

the perioperative patient data dimension reduction device is used for performing dimension reduction processing on the original perioperative characteristic data of all patients to obtain corresponding perioperative characteristic data;

the sample data set acquisition module is used for associating a classification label set corresponding to the original perioperative period characteristic data for the sample by taking perioperative period characteristic data of the patient as the sample to obtain a sample data set of the perioperative period patient;

further comprising perioperative patient sample data set equalisation means as claimed in claim 8 for equalising the sample data set.

10. The perioperative patient sample dataset acquisition system of claim 9, further comprising a missing filling means for filling missing values in the original perioperative feature data of the patient and inputting the filled original perioperative feature data into the perioperative patient data dimension reduction means for performing dimension reduction.

Images