Financial distress prediction method and device based on semi-supervised discriminant analysis, equipment and storage medium

Abstract

The application discloses a financial distress prediction method and device based on semi-supervised discriminant analysis, equipment and storage medium, the method comprises the following steps: the initial pseudo label is given to the unlabeled data set based on the ownership discriminant model constructed based on the labeled data set, the TDEA-DA model containing the base classifier of multiple scale returns is trained based on the labeled data set and the unlabeled data set after the initial pseudo label is given, a target prediction model is obtained, and the financial distress prediction of the to-be-predicted financial sample is carried out based on the target prediction model; the application gives the pseudo label to the unlabeled data by fully utilizing the limited labeled data, trains the TDEA-DA model containing the base classifier of multiple scale returns by using the labeled sample and the unlabeled sample with the pseudo label, so that the influence of different scale return assumptions on the financial distress prediction is considered, which is more in line with the actual situation, and the accuracy and stability of the financial distress prediction are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of financial risk prediction, and in particular to a method, apparatus, device, and storage medium for predicting financial distress based on semi-supervised discriminant analysis. Background Technology

[0002] Financial distress is the result of accumulated financial risks. In a broad sense, financial distress includes not only insolvency and default, but also bankruptcy proceedings, operational disruptions, and regulatory interventions such as "Special Treatment" (ST) or delisting risks. Financial distress has become a core risk factor threatening the long-term survival and sustainable development of enterprises. Therefore, improving the accuracy and reliability of predicting financial risks and financial distress is a pressing issue in the financial sector.

[0003] In recent years, with the significant improvement in computing power, machine learning methods have made remarkable progress and have become a key tool for predicting financial distress. However, traditional machine learning methods, such as support vector machines in supervised learning, typically rely on large amounts of labeled data to train models to ensure their accuracy and generalization ability. However, in the financial field, obtaining high-quality labeled data is often costly and time-consuming. For example, the low frequency of financial distress events and the limited transparency of early-stage financial distress lead to a scarcity of labeled data, which greatly limits the application of traditional supervised learning methods. Therefore, in scenarios targeting the prediction of financial risk and financial distress in the financial sector, the high cost of acquiring labeled data and the scarcity of data resources result in insufficient generalization ability of traditional supervised learning models, leading to low prediction accuracy and reliability. Summary of the Invention

[0004] The main objective of this invention is to provide a financial distress prediction method, apparatus, device, and storage medium based on semi-supervised discriminant analysis. This invention aims to solve the problems of overfitting and insufficient generalization ability of models caused by data scarcity in existing supervised learning technologies. These problems prevent accurate prediction of financial distress in complex financial scenarios, thus hindering users from responding to financial risks in a timely manner.

[0005] To achieve the above objectives, this invention provides a financial distress prediction method based on semi-supervised discriminant analysis, the method comprising the following steps:

[0006] Obtain a financial ratio dataset, which includes a training dataset, comprising a labeled dataset and an unlabeled dataset, wherein the labeled dataset includes labeled samples of a first category and labeled samples of a second category.

[0007] The unlabeled dataset is input into the attribution discrimination model, which outputs the discrimination result. Based on the discrimination result, the unlabeled dataset is assigned an initial pseudo-label. The unlabeled dataset after being assigned the initial pseudo-label includes a first category pseudo-label sample and a second category pseudo-label sample. The attribution discrimination model is a DEA-DA model built based on multiple scale rewards. The attribution discrimination model is trained on the labeled dataset. The scale rewards include unlimited scale rewards, increasing scale rewards, decreasing scale rewards, and constant scale rewards.

[0008] The TDEA-DA model is trained based on the labeled dataset and the unlabeled dataset after being given initial pseudo-labels. The TDEA-DA model includes base classifiers corresponding to various scale rewards.

[0009] In response to the training TDEA-DA model not satisfying the commutation condition, the target function values ​​corresponding to each base classifier are output.

[0010] The TDEA-DA model is adjusted based on the objective function value to obtain the target prediction model, and financial distress prediction is performed on the financial sample to be predicted based on the target prediction model.

[0011] Optionally, training the TDEA-DA model based on the labeled dataset and the unlabeled dataset with initial pseudo-labels includes:

[0012] Initialize the training parameters, which include the sample class weights and model parameters of the training dataset of the TDEA-DA model;

[0013] The initialized training parameters, the labeled dataset, and the unlabeled dataset with initial pseudo-labels are input into the TDEA-DA model, and the first stage optimization of the TDEA-DA model is performed, outputting the model parameters and result sample set after the first stage optimization.

[0014] Determine whether the resulting sample set is linearly separable;

[0015] If the result sample set is linearly inseparable, then the second stage optimization of the TDEA-DA model is performed based on the result sample set and the model parameters optimized in the first stage, and the model parameters optimized in the second stage are output.

[0016] The first category attribution strategy is determined based on the model parameters optimized in the second stage.

[0017] Based on the first category attribution discrimination strategy, the samples in the training dataset of the TDEA-DA model are classified to obtain the trained TDEA-DA model;

[0018] Determine whether the trained TDEA-DA model satisfies the commutation condition;

[0019] If the trained TDEA-DA model satisfies the exchange condition, then the pseudo-labels of the pseudo-labeled sample pairs that satisfy the exchange condition are swapped, and the process returns to the first stage of optimization of the TDEA-DA model.

[0020] If the trained TDEA-DA model does not meet the exchange condition, then the objective function values ​​corresponding to each base classifier are output, and the TDEA-DA model is adjusted based on the objective function values ​​to obtain the target prediction model.

[0021] Optionally, after determining whether the result sample set is linearly separable, the method further includes:

[0022] If the result sample set is linearly separable, then the second category classification strategy is determined based on the model parameters optimized in the first stage.

[0023] Based on the second category classification strategy, each sample in the training dataset of the TDEA-DA model is classified to obtain the trained TDEA-DA model, and the step of determining whether the trained TDEA-DA model satisfies the exchange condition is executed.

[0024] Optionally, the optimized model parameters in the first stage include the optimized weights of the input features, the weights of the output features, and the classification threshold parameters; the optimized model parameters in the second stage include the optimized classification boundary parameters, the weights of the input features, and the weights of the output features in the linearly inseparable scenario.

[0025] The first category attribution strategy includes:

[0026]

[0027] in, This represents the test samples in the test set. Indicates test sample The Middle One input characteristic, Indicates test sample The Middle Individual output characteristics This represents the total number of input-type characteristics. This represents the total number of output-type characteristics. This represents the weight of the input features output after the second stage of optimization. This represents the weight of the output feature after the second stage of optimization. This represents the classification boundary parameters output after the second stage of optimization. This indicates a labeled dataset. D The first category contains labeled samples. This indicates a labeled dataset. D The second category contains labeled samples;

[0028] The second category attribution strategy includes:

[0029]

[0030] in, This represents the weights of the input features output after the first stage of optimization. This represents the weight of the output features after the first stage of optimization. This represents the classification threshold parameter output after the first stage of optimization. This represents the interval of the discriminant function.

[0031] Optionally, the exchange conditions include:

[0032] Does the training dataset of the TDEA-DA model contain a pair of unlabeled samples with different pseudo-labels that satisfy the following condition:

[0033]

[0034]

[0035] in, This represents the first category of pseudo-labeled samples. Positive slack variables, For the second category of pseudo-label samples Negative slack variables, This represents an unlabeled dataset. This represents the weight of the pseudo-labeled sample in the first category. This represents the weight of the pseudo-labeled sample in the second category.

[0036] Optionally, the first stage of optimization includes:

[0037] Model parameters are optimized using a first objective function and a first constraint condition, wherein the first objective function includes:

[0038]

[0039] in, This indicates a labeled dataset. D The first category contains labeled samples. This indicates a labeled dataset. D The second category contains labeled samples. Represents unlabeled datasets UThe first category of pseudo-labeled samples, Represents unlabeled datasets U The second category of pseudo-label samples, This represents the positive slack variable for labeled samples in the first category. This represents the negative slack variable for labeled samples in the second category. Indicates the sample class weights. This represents the weight of the pseudo-labeled sample in the first category. The positive relaxation variable represents the pseudo-labeled samples of the first category. This represents the weight of the pseudo-labeled samples in the second category. The negative slack variable represents the pseudo-labeled samples of the second category;

[0040] The first constraint includes:

[0041]

[0042]

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049] in, Indicates input-type characteristics, To input feature index, For sample index, Indicating input characteristics The weight, Indicates output-type characteristics, Indicates the output feature index. Indicates output characteristics The weight, This represents the classification threshold parameter. This represents a positive tolerance parameter, which provides tolerance space for constraint differences among labeled samples. , and Constraint variables representing returns to scale. This represents the negative slack variable for labeled samples in the first category. This represents the positive relaxation variable for labeled samples in the second category. This represents the negative slack variable for the first category of pseudo-labeled samples. The positive relaxation variable represents the pseudo-labeled samples of the second category. It is a positive real number;

[0050] The result sample set includes:

[0051]

[0052]

[0053]

[0054]

[0055]

[0056]

[0057]

[0058]

[0059]

[0060] in, Represents the result sample set, , , and They represent in , , and The set of samples that were not correctly assigned to a specific category. This indicates that the discriminant function generated by the model classifies it as belonging to... There are categorized labeled samples. This indicates that the discriminant function generated by the model classifies it as belonging to... There are categorized labeled samples. This indicates that the discriminant function generated by the model classifies it as belonging to... Unlabeled samples of each category This indicates that the discriminant function generated by the model classifies it as belonging to... Unlabeled samples of each category This represents the weights of the input features output after the first stage of optimization. This represents the weight of the output features after the first stage of optimization. This represents the classification threshold parameter output after the first stage of optimization;

[0061] The second phase of optimization includes:

[0062] The model parameters are optimized using a second objective function and a second constraint condition. The second objective function includes:

[0063]

[0064] The second constraint includes:

[0065]

[0066]

[0067]

[0068]

[0069]

[0070]

[0071] in, This represents the classification boundary parameters in linearly inseparable scenarios.

[0072] Optionally, adjusting the TDEA-DA model based on the objective function value to obtain a target prediction model, and then predicting financial distress for the financial sample to be predicted based on the target prediction model, includes:

[0073] Based on the objective function value, the target weight values ​​of each base classifier in the TDEA-DA model are determined and adjusted, referring to the following formula:

[0074]

[0075] in, Indicates the first The target weight values ​​of each base classifier Indicates the first The target function value output by each base classifier This represents the sum of the reciprocals of the objective function values ​​of all base classifiers;

[0076] The TDEA-DA model is adjusted based on the target weight values ​​to obtain the target prediction model, and financial distress prediction is performed on the financial sample to be predicted based on the target prediction model.

[0077] Furthermore, to achieve the above objectives, the present invention also proposes a financial distress prediction device based on semi-supervised discriminant analysis, wherein the financial distress prediction device based on semi-supervised discriminant analysis includes:

[0078] The dataset acquisition module is used to acquire a financial ratio dataset, which includes a training dataset, a labeled dataset and an unlabeled dataset, and the labeled dataset includes labeled samples of a first category and labeled samples of a second category.

[0079] The pseudo-label assignment module is used to input the unlabeled dataset into the attribution discrimination model, output the discrimination result, and assign initial pseudo-labels to the unlabeled dataset based on the discrimination result. The unlabeled dataset after being assigned initial pseudo-labels includes first-class pseudo-label samples and second-class pseudo-label samples. The attribution discrimination model is a DEA-DA model built based on multiple scale rewards. The attribution discrimination model is trained based on the labeled dataset. The scale rewards include unlimited scale rewards, increasing scale rewards, decreasing scale rewards, and constant scale rewards.

[0080] The model training module is used to train the TDEA-DA model based on the labeled dataset and the unlabeled dataset after assigning initial pseudo-labels. The TDEA-DA model includes base classifiers corresponding to various scale rewards.

[0081] The output response module is used to output the objective function values ​​of each base classifier in response to the trained TDEA-DA model not satisfying the commutation condition.

[0082] The financial distress prediction module is used to adjust the TDEA-DA model based on the objective function value to obtain the target prediction model, and to predict the financial distress of the financial sample to be predicted based on the target prediction model.

[0083] Furthermore, to achieve the above objectives, this application also proposes a financial distress prediction device based on semi-supervised discriminant analysis. The device includes: a memory, a processor, and a financial distress prediction program based on semi-supervised discriminant analysis stored in the memory. The processor is used to run the financial distress prediction program based on semi-supervised discriminant analysis, and the computer program is configured to implement the steps of the financial distress prediction method based on semi-supervised discriminant analysis as described above.

[0084] In addition, to achieve the above objectives, this application also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the financial distress prediction method based on semi-supervised discriminant analysis as described above.

[0085] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the financial distress prediction method based on semi-supervised discriminant analysis as described above.

[0086] This invention acquires a financial ratio dataset, which includes a training dataset comprising a labeled dataset and an unlabeled dataset. The labeled dataset includes labeled samples from a first class and labeled samples from a second class. The unlabeled dataset is input into a classification discriminant model, which outputs a classification result. Based on the classification result, initial pseudo-labels are assigned to the unlabeled dataset. The unlabeled dataset after initial pseudo-labeling includes pseudo-labeled samples from the first class and pseudo-labeled samples from the second class. The classification discriminant model is a DEA-DA model based on multiple scale rewards. The classification discriminant model is trained on the labeled dataset, and the scale rewards include unrestricted scale rewards, increasing scale rewards, decreasing scale rewards, and constant scale rewards. The TDEA-DA model is trained on the labeled dataset and the unlabeled dataset after initial pseudo-labeling. The TDEA-DA model includes various... The invention employs a base classifier corresponding to returns to scale. In response to the trained TDEA-DA model failing to meet the commutative condition, it outputs the objective function values ​​corresponding to each base classifier. Based on the objective function values, the TDEA-DA model is adjusted to obtain the target prediction model, and financial distress prediction is performed on the financial sample to be predicted based on the target prediction model. Because this invention uses a DEA-DA model constructed based on multiple returns to scale, combined with labeled datasets to assign initial pseudo-labels to unlabeled datasets, it fully utilizes limited labeled data to assign pseudo-labels to unlabeled data. The TDEA-DA model, containing base classifiers with multiple returns to scale, is trained using labeled samples and unlabeled samples with pseudo-labels. This approach considers the impact of different returns to scale assumptions on financial distress prediction, ensuring that the prediction results better reflect the actual situation of the complex financial environment, effectively improving the accuracy and stability of financial distress prediction. Attached Figure Description

[0087] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0088] Figure 1 This is a schematic diagram of the structure of a financial distress prediction device based on semi-supervised discriminant analysis in the hardware operating environment of the embodiment of the present invention;

[0089] Figure 2 This is a flowchart illustrating an embodiment of the financial distress prediction method based on semi-supervised discriminant analysis of the present invention.

[0090] Figure 3 This is a schematic diagram of the model iterative update process in one embodiment of the financial distress prediction method based on semi-supervised discriminant analysis of the present invention;

[0091] Figure 4 This is a schematic diagram of the framework of the TDEA-DA model in one embodiment of the financial distress prediction method based on semi-supervised discriminant analysis of the present invention;

[0092] Figure 5 This is a structural block diagram of an embodiment of the financial distress prediction device based on semi-supervised discriminant analysis of the present invention.

[0093] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0094] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0095] Reference Figure 1 , Figure 1 This is a schematic diagram of the structure of a financial distress prediction device based on semi-supervised discriminant analysis in the hardware operating environment of the embodiment of the present invention.

[0096] like Figure 1 As shown, the financial distress prediction device based on semi-supervised discriminant analysis may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The memory 1005 may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0097] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on financial distress prediction devices based on semi-supervised discriminant analysis and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0098] like Figure 1As shown, the memory 1005, which is a computer-readable storage medium, may include an operating system, a network communication module, a user interface module, and a financial distress prediction program based on semi-supervised discriminant analysis.

[0099] exist Figure 1 In the financial distress prediction device based on semi-supervised discriminant analysis shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and memory 1005 in the financial distress prediction device based on semi-supervised discriminant analysis of the present invention can be set in the financial distress prediction device based on semi-supervised discriminant analysis. The financial distress prediction device based on semi-supervised discriminant analysis calls the financial distress prediction program based on semi-supervised discriminant analysis stored in the memory 1005 through the processor 1001 and executes the financial distress prediction method based on semi-supervised discriminant analysis provided in the embodiments of the present invention.

[0100] This invention provides a method for predicting financial distress based on semi-supervised discriminant analysis, referring to... Figure 2 , Figure 2 This is a flowchart illustrating an embodiment of the financial distress prediction method based on semi-supervised discriminant analysis of the present invention.

[0101] In this embodiment, the financial distress prediction method based on semi-supervised discriminant analysis includes the following steps:

[0102] Step S10: Obtain the financial ratio dataset.

[0103] It should be noted that this embodiment is applied to the prediction of financial risk and financial distress in the financial field. This embodiment analyzes the heterogeneity of financial data under different returns to scale assumptions. First, it initializes pseudo-labels for all unlabeled samples using an improved DEA-DA model based on different returns to scale assumptions. Then, it uses TDEA-DA based on the improved DEA-DA model under different returns to scale assumptions as a base classifier to discriminate samples in all training sets. During this process, a designed pseudo-label update strategy is incorporated, using a pseudo-label algorithm to iteratively update the pseudo-labels to improve the model's efficiency and accuracy in utilizing unlabeled data. Finally, based on the objective function values ​​of each base classifier, the TDEA-DA discrimination results under different assumptions are weighted and integrated through voting to balance the prediction results of different models, reduce the impact of single model bias on the overall evaluation, and thus form a target prediction model for predicting future financial distress.

[0104] It should be understood that the executing entity of this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or a terminal electronic device capable of performing the above functions. The following description uses a financial distress prediction device based on semi-supervised discriminant analysis (hereinafter referred to as the prediction device) as an example to illustrate this embodiment and the following embodiments.

[0105] It should be noted that the financial ratio dataset includes a training dataset, which includes a labeled dataset and an unlabeled dataset. The labeled dataset includes labeled samples of a first category and labeled samples of a second category.

[0106] It should be noted that the financial ratio dataset contains a set of corporate financial indicators (such as liquidity, profitability, and solvency ratios) used to quantify a company's financial condition. Labeled datasets can be subsets of data where samples are accompanied by true category labels (such as "healthy" / "distressed"), used for supervised learning. Unlabeled datasets can be subsets of data containing only financial features but without category labels, reflecting unlabeled samples in real-world scenarios.

[0107] In some embodiments, the first category of labeled samples may be samples of financial health labels (e.g., three consecutive years of profitability, debt ratio <50%), and the second category of labeled samples may be samples of financial distress labels (e.g., bankruptcy, ST treatment, or continuous losses).

[0108] In some embodiments, the prediction device may collect raw financial data from a financial database and then preprocess the raw financial data, such as performing data cleaning (to improve data quality and reduce the impact of abnormal quality data) and data standardization (to eliminate dimensional differences) to obtain a financial ratio dataset.

[0109] Step S20: Input the unlabeled dataset into the attribution discrimination model, output the discrimination result, and assign initial pseudo-labels to the unlabeled dataset based on the discrimination result.

[0110] It should be noted that the unlabeled dataset after assigning initial pseudo-labels includes pseudo-labeled samples of the first category and pseudo-labeled samples of the second category. The attribution discrimination model is a DEA-DA model built based on multiple scale rewards. The attribution discrimination model is trained on the labeled dataset. The scale rewards include unrestricted scale rewards, increasing scale rewards, decreasing scale rewards, and invariant scale rewards.

[0111] It should be noted that unlimited returns to scale (VRS) means that the elasticity of input and output is variable, which is suitable for non-equilibrium financial structures; increasing returns to scale (IRS) means that efficiency increases as scale increases, which is suitable for growth companies; decreasing returns to scale (DRS) means that efficiency decreases as scale increases, which is suitable for mature companies; and constant returns to scale (CRS) means that scale and efficiency are proportional, which is suitable for companies with stable operations.

[0112] It should be noted that the DEA-DA model is a model constructed by integrating the frameworks of Data Envelopment Analysis (DEA) and Discriminant Analysis (DA). By introducing the DEA model into the framework of Discriminant Analysis to calculate the weight of each feature (variable), it is used to classify new samples. It has a non-parametric framework and strong adaptability.

[0113] It is understandable that this embodiment addresses the practical scenario of financial distress prediction. Considering the impact of a large amount of unlabeled data on model performance in real-world applications, it proposes a semi-supervised learning algorithm based on an improved DEA-DA framework. This algorithm, through pseudo-label initialization and iterative refinement mechanisms, fully utilizes both limited labeled data and a large amount of unlabeled data, effectively improving the accuracy and stability of financial distress prediction.

[0114] In some embodiments, the prediction device can train independent DEA-DA models for four scale rewards: unrestricted (VRS), increasing (IRS), decreasing (DRS), and invariant (CRS). Each DEA-DA model takes a labeled dataset as input, learns a discrimination boundary, and outputs the probability of sample classification. Unlabeled samples are input into the DEA-DA models with different scale rewards, outputting a first-class pseudo-labeled sample (predicted as healthy) and a second-class pseudo-labeled sample (predicted as distressed). The initial pseudo-labels are used for initial guidance in semi-supervised learning.

[0115] Step S30: Train the TDEA-DA model based on the labeled dataset and the unlabeled dataset after assigning initial pseudo-labels.

[0116] It should be noted that the TDEA-DA model is an improved and optimized version of the DEDA-DA model based on a semi-supervised learning framework, and it can use both labeled and pseudo-labeled samples simultaneously.

[0117] The TDEA-DA model includes base classifiers for each type of scale return. For example, the TDEA-DA model includes base classifiers for unrestricted scale returns, increasing scale returns, decreasing scale returns, and constant scale returns.

[0118] In the specific implementation, the prediction device integrates the labeled dataset with the unlabeled dataset after assigning initial pseudo-labels and the training dataset, trains a TDEA-DA base classifier for each class of scale reward (VRS / IRS / DRS / CRS), constructs the TDEA-DA model, inputs the training dataset into the TDEA-DA model, and iteratively trains each base classifier.

[0119] In some embodiments, the financial ratio dataset includes not only the training dataset but also a test set. The test set is used to test the trained TDEA-DA model to determine whether the trained TDEA-DA model meets the conditions. If the conditions are not met, training continues. If the preset conditions are met, the trained TDEA-DA model is output as the target prediction model.

[0120] Furthermore, to improve the accuracy of model predictions, step S30 above may include:

[0121] Step S301: Initialize training parameters, which include the sample class weights and model parameters of the training dataset of the TDEA-DA model;

[0122] Step S302: Input the initialized training parameters, the labeled dataset, and the unlabeled dataset with initial pseudo-labels into the TDEA-DA model, and perform the first stage optimization of the TDEA-DA model, outputting the model parameters and result sample set after the first stage optimization;

[0123] Step S303: Determine whether the result sample set is linearly separable;

[0124] Step S304: If the result sample set is linearly inseparable, then perform the second stage optimization of the TDEA-DA model based on the result sample set and the model parameters optimized in the first stage, and output the model parameters optimized in the second stage;

[0125] Step S305: Determine the first category classification strategy based on the optimized model parameters from the second stage;

[0126] Step S306: Assign categories to each sample in the training dataset of the TDEA-DA model based on the first category attribution discrimination strategy to obtain the trained TDEA-DA model;

[0127] Step S307: Determine whether the trained TDEA-DA model satisfies the commutation condition;

[0128] Step S308: If the trained TDEA-DA model satisfies the exchange condition, then perform pseudo-label exchange on the pseudo-label sample pairs that satisfy the exchange condition, and return to the step of performing the first stage optimization of the TDEA-DA model.

[0129] Step S309: If the trained TDEA-DA model does not meet the exchange condition, then execute the output of the objective function values ​​corresponding to each base classifier, and adjust the TDEA-DA model based on the objective function values ​​to obtain the target prediction model.

[0130] It should be noted that the optimized model parameters in the first stage include the optimized weights of the input features, the weights of the output features, and the classification threshold parameters; the optimized model parameters in the second stage include the optimized classification boundary parameters, the weights of the input features, and the weights of the output features in the linearly inseparable scenario.

[0131] In its implementation, the prediction device iteratively trains the TDEA-DA base classifier across all scale rewards. The iterative training process is as follows:

[0132] Step 1: Initialize sample class weights , , Weights and discriminant function margins of two classes of unlabeled samples .in for Medium sample size and The ratio of the sample size to the median sample size , usually set as , ; It is a very small positive number greater than zero. The value is set according to the actual situation, and is usually set as follows: .

[0133] Step 2: Run the first stage of TDEA-DA optimization, and then determine whether the data is linearly separable based on the output of the first stage optimization. If it is not linearly separable, proceed to step 3. If it is linearly separable, directly use the optimization model parameters of the first stage to execute step 6.

[0134] The first phase of optimization includes:

[0135] Model parameters are optimized using a first objective function and a first constraint condition, wherein the first objective function includes:

[0136]

[0137] in, This indicates a labeled dataset. D The first category contains labeled samples. This indicates a labeled dataset.D The second category contains labeled samples. Represents unlabeled datasets U The first category of pseudo-labeled samples, Represents unlabeled datasets U The second category of pseudo-label samples, This represents the positive slack variable for labeled samples in the first category. This represents the negative slack variable for labeled samples in the second category. Indicates the sample class weights. This represents the weight of the pseudo-labeled sample in the first category. The positive relaxation variable represents the pseudo-labeled samples of the first category. This represents the weight of the pseudo-labeled samples in the second category. The negative slack variable represents the pseudo-labeled samples of the second category;

[0138] The first constraint includes:

[0139]

[0140]

[0141]

[0142]

[0143]

[0144]

[0145]

[0146]

[0147] in, Indicates input-type characteristics, To input feature index, For sample index, Indicates input characteristics The weight, Indicates output-type characteristics, Indicates the output feature index. Indicates output characteristics The weight, This represents the classification threshold parameter. This represents a positive tolerance parameter, which provides tolerance space for constraint differences among labeled samples. , and Constraint variables representing returns to scale. This represents the negative slack variable for labeled samples in the first category. This represents the positive relaxation variable for labeled samples in the second category. This represents the negative slack variable for the first category of pseudo-labeled samples. The positive relaxation variable represents the pseudo-labeled samples of the second category. It is a positive real number;

[0148] The optimal solution obtained from the first stage of optimization ( , , This will yield the following sample set:

[0149]

[0150]

[0151]

[0152]

[0153]

[0154]

[0155]

[0156]

[0157]

[0158] in, Represents the result sample set, , , and They represent in , , and The set of samples that were not correctly assigned to a specific category. This indicates that the discriminant function generated by the model classifies it as belonging to... There are categorized labeled samples. This indicates that the discriminant function generated by the model classifies it as belonging to... There are categorized labeled samples. This indicates that the discriminant function generated by the model classifies it as belonging to... Unlabeled samples of each category This indicates that the discriminant function generated by the model classifies it as belonging to... Unlabeled samples of each category This represents the weights of the input features output after the first stage of optimization. This represents the weight of the output features after the first stage of optimization. This represents the classification threshold parameter output after the first stage of optimization.

[0159] Step 3: Perform the second stage optimization of TDEA-DA and output the optimized model parameters. The second stage optimization includes:

[0160] The model parameters are optimized using a second objective function and a second constraint condition. The second objective function includes:

[0161]

[0162] The second constraint includes:

[0163]

[0164]

[0165]

[0166]

[0167]

[0168]

[0169] in, This represents the classification boundary parameters in linearly inseparable scenarios.

[0170] Step 4: Based on the optimal solution obtained in the second stage of optimization ( , , Use the following first-class classification strategy to determine the classification of any sample. The first category classification strategy includes:

[0171]

[0172] in, This represents the test samples in the test set. Indicates test sample The Middle One input characteristic, Indicates test sample The Middle Individual output characteristics This represents the total number of input-type characteristics. This represents the total number of output-type characteristics. This represents the weight of the input features output after the second stage of optimization. This represents the weight of the output feature after the second stage of optimization. This represents the classification boundary parameters output after the second stage of optimization. This indicates a labeled dataset. D The first category contains labeled samples. This indicates a labeled dataset. D The second category contains labeled samples.

[0173] Step 5: Iterative Update. Whether an iterative update is needed can be determined by... Figure 3 Make a judgment. Figure 3 This is a schematic diagram of the model iterative update process, by Figure 3 It can be seen that if there exists a pair of unlabeled samples assigned different pseudo-labels... satisfy:

[0174]

[0175] and , If it is greater than 1, then let:

[0176]

[0177] Then swap the pseudo-labels of the sample pairs that satisfy the slack variables and the largest pseudo-label in the swap condition, and return to sub-step two to run TDEA-DA again for the next round of iteration. If there is no such pseudo-label sample pair or the number of iterations reaches the upper limit set by the user, the iteration stops.

[0178] Step Six: If there are no samples available for performing the second stage optimization of TDEA-DA, i.e., the result sample set... The fact that the set is empty indicates that the dataset is linearly separable, so we can directly use the optimal solution from the first stage. , , Use the following second-category classification strategy to determine the classification of any sample. Category classification:

[0179]

[0180] in, This represents the weights of the input features output after the first stage of optimization. This represents the weight of the output features after the first stage of optimization. This represents the classification threshold parameter output after the first stage of optimization. This represents the interval of the discriminant function.

[0181] Furthermore, in order to shorten the decision-making cycle for predicting financial distress, the following steps are included after step S303:

[0182] Step S3031: If the result sample set is linearly separable, then determine the second category classification strategy based on the model parameters optimized in the first stage;

[0183] Step S3032: Based on the second category attribution discrimination strategy, classify each sample in the training dataset of the TDEA-DA model to obtain the trained TDEA-DA model, and execute the step of judging whether the trained TDEA-DA model satisfies the exchange condition.

[0184] It should be noted that if there are no samples available for performing the second-stage optimization of TDEA-DA, i.e., the result sample set... The fact that the set is empty indicates that the dataset is linearly separable, so we can directly use the optimal solution from the first stage. , , The second-class classification strategy is used to determine the classification of any sample. Category attribution.

[0185] Step S40: In response to the trained TDEA-DA model not satisfying the commutation condition, output the objective function values ​​corresponding to each base classifier.

[0186] It should be noted that the commutation condition can serve as a stopping mechanism for the iterative training of the TDEA-DA model, based on the slack variable threshold of the pseudo-label sample pairs to ensure the rationality of pseudo-label adjustments. When the TDEA-DA model satisfies the commutation condition, it is determined that the model has optimization space; when the TDEA-DA model does not satisfy the commutation condition, it is determined that the model has been trained successfully and can be used for financial distress prediction.

[0187] The objective function value can be the solution to the optimization problem of the base classifier in the TDEA-DA model, reflecting the performance of the base classifier (the smaller the value, the better the classification boundary).

[0188] It is understood that this embodiment improves computational efficiency by iteratively training the TDEA-DA model, avoiding invalid iterations by exchanging conditions, and updating the model only when there is room for optimization in the pseudo-labels; the objective function value provides an objective indicator for subsequent weight allocation, replacing subjective experience judgments and improving prediction accuracy.

[0189] Step S50: Adjust the TDEA-DA model based on the objective function value to obtain the target prediction model, and predict the financial distress of the financial sample to be predicted based on the target prediction model.

[0190] It is understandable that this embodiment takes into account the heterogeneity of financial data under different scale return assumptions. When predicting financial distress, it introduces four TDEA-DA base classifiers under different scale return assumptions and describes the problem from a global perspective through weighted integration. At the same time, it uses a pseudo-label update exchange strategy to dynamically update sample labels and explore the impact of different scale return assumptions on financial distress prediction. This approach is more in line with reality and has better generalization ability and adaptability.

[0191] Furthermore, to reduce the cost of financial risk assessment and improve its quality, step S50 above may include:

[0192] Step S501: Based on the objective function value, determine and adjust the target weight values ​​of each base classifier in the TDEA-DA model.

[0193] Understandably, the prediction device uses the objective function value of TDEA-DA in each case during the last iteration to derive the target weight value of each base classifier in the final target prediction model. Refer to the following formula:

[0194]

[0195] in, Indicates the first The target weight values ​​of each base classifier Indicates the first The target function value output by each base classifier This represents the sum of the reciprocals of the objective function values ​​of all base classifiers.

[0196] Step S502: Adjust the TDEA-DA model based on the target weight value to obtain the target prediction model, and predict the financial distress of the financial sample to be predicted based on the target prediction model.

[0197] In the specific implementation, refer to Figure 4 , Figure 4 This is a schematic diagram of the TDEA-DA model framework. The TDEA-DA model includes TDEA-DA1, TDEA-DA2, TDEA-DA3, and TDEA-DA4, corresponding to four different scale rewards. Each base classifier is trained using a pseudo-label exchange strategy to obtain the final base classifier. Finally, a weighted voting method is used to select the appropriate base classifier for any test set. Samples in The TDEA-DA base classifiers trained for four different scenarios are used to determine the category attribution in order to achieve the goal of predicting financial distress.

[0198] This embodiment acquires a financial ratio dataset, which includes a training dataset comprising a labeled dataset and an unlabeled dataset. The labeled dataset includes labeled samples from a first category and labeled samples from a second category. The unlabeled dataset is input into a classification model, which outputs a classification result. Based on the classification result, initial pseudo-labels are assigned to each unlabeled sample in the unlabeled dataset. The unlabeled dataset after assigning initial pseudo-labels includes pseudo-labeled samples from the first and second categories. The classification model is a DEA-DA model built based on multiple scale rewards. The classification model is trained on the labeled dataset, and the scale rewards include unrestricted scale rewards, increasing scale rewards, decreasing scale rewards, and constant scale rewards. The TDEA-DA model is trained on the labeled dataset and the unlabeled dataset after assigning initial pseudo-labels. The model includes base classifiers corresponding to various returns to scale. In response to the trained TDEA-DA model not satisfying the commutative condition, it outputs the objective function values ​​corresponding to each base classifier. Based on the objective function values, the TDEA-DA model is adjusted to obtain the target prediction model, and financial distress prediction is performed on the financial sample to be predicted based on the target prediction model. Because this invention uses a DEA-DA model constructed based on multiple returns to scale, combined with labeled datasets to assign initial pseudo-labels to unlabeled datasets, it fully utilizes limited labeled data to assign pseudo-labels to unlabeled data. The TDEA-DA model, containing base classifiers with multiple returns to scale, is trained using labeled samples and unlabeled samples with pseudo-labels. This considers the impact of different returns to scale assumptions on financial distress prediction, ensuring that the prediction results are more consistent with the actual situation of the complex financial environment, effectively improving the accuracy and stability of financial distress prediction.

[0199] Furthermore, embodiments of the present invention also propose a computer-readable storage medium storing a financial distress prediction program based on semi-supervised discriminant analysis. When the financial distress prediction program based on semi-supervised discriminant analysis is executed by a processor, it implements the steps of the financial distress prediction method based on semi-supervised discriminant analysis as described above.

[0200] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0201] The aforementioned computer-readable storage medium may be included in a financial distress prediction device based on semi-supervised discriminant analysis; or it may exist independently and not assembled into a financial distress prediction device based on semi-supervised discriminant analysis.

[0202] Furthermore, this invention also proposes a computer program product, including a financial distress prediction program based on semi-supervised discriminant analysis, wherein when the financial distress prediction program based on semi-supervised discriminant analysis is executed by a processor, it implements the steps of the financial distress prediction method based on semi-supervised discriminant analysis as described above.

[0203] The specific implementation of the computer program product of the present invention is basically the same as the embodiments of the financial distress prediction method based on semi-supervised discriminant analysis described above, and will not be repeated here.

[0204] Reference Figure 5 , Figure 5 This is a structural block diagram of an embodiment of the financial distress prediction device based on semi-supervised discriminant analysis of the present invention.

[0205] like Figure 5 As shown, the financial distress prediction device based on semi-supervised discriminant analysis proposed in this embodiment of the invention includes:

[0206] The dataset acquisition module 10 is used to acquire a financial ratio dataset, which includes a training dataset, which includes a labeled dataset and an unlabeled dataset. The labeled dataset includes labeled samples of a first category and labeled samples of a second category.

[0207] The pseudo-label assignment module 20 is used to input the unlabeled dataset into the attribution discrimination model, output the discrimination result, and assign initial pseudo-labels to the unlabeled dataset based on the discrimination result. The unlabeled dataset after being assigned initial pseudo-labels includes first-class pseudo-label samples and second-class pseudo-label samples. The attribution discrimination model is a DEA-DA model built based on multiple scale rewards. The attribution discrimination model is trained based on the labeled dataset. The scale rewards include unlimited scale rewards, increasing scale rewards, decreasing scale rewards, and constant scale rewards.

[0208] The model training module 30 is used to train the TDEA-DA model based on the labeled dataset and the unlabeled dataset after assigning initial pseudo-labels. The TDEA-DA model includes base classifiers corresponding to various scale rewards.

[0209] The output response module 40 is used to output the objective function values ​​of each base classifier in response to the trained TDEA-DA model not satisfying the commutation condition.

[0210] The financial distress prediction module 50 is used to adjust the TDEA-DA model based on the objective function value to obtain the target prediction model, and to predict the financial distress of the financial sample to be predicted based on the target prediction model.

[0211] This embodiment acquires a financial ratio dataset, which includes a training dataset comprising a labeled dataset and an unlabeled dataset. The labeled dataset includes labeled samples from a first category and labeled samples from a second category. The unlabeled dataset is input into a classification model, which outputs a classification result. Based on the classification result, initial pseudo-labels are assigned to each unlabeled sample in the unlabeled dataset. The unlabeled dataset after assigning initial pseudo-labels includes pseudo-labeled samples from the first and second categories. The classification model is a DEA-DA model built based on multiple scale rewards. The classification model is trained on the labeled dataset, and the scale rewards include unrestricted scale rewards, increasing scale rewards, decreasing scale rewards, and constant scale rewards. The TDEA-DA model is trained on the labeled dataset and the unlabeled dataset after assigning initial pseudo-labels. The model includes base classifiers corresponding to various returns to scale. In response to the trained TDEA-DA model not satisfying the commutative condition, it outputs the objective function values ​​corresponding to each base classifier. Based on the objective function values, the TDEA-DA model is adjusted to obtain the target prediction model, and financial distress prediction is performed on the financial sample to be predicted based on the target prediction model. Because this invention uses a DEA-DA model constructed based on multiple returns to scale, combined with labeled datasets to assign initial pseudo-labels to unlabeled datasets, it fully utilizes limited labeled data to assign pseudo-labels to unlabeled data. The TDEA-DA model, containing base classifiers with multiple returns to scale, is trained using labeled samples and unlabeled samples with pseudo-labels. This considers the impact of different returns to scale assumptions on financial distress prediction, ensuring that the prediction results are more consistent with the actual situation of the complex financial environment, effectively improving the accuracy and stability of financial distress prediction.

[0212] The financial distress prediction device based on semi-supervised discriminant analysis provided in this application employs the financial distress prediction method based on semi-supervised discriminant analysis in the above embodiments, and can solve the technical problems of financial distress prediction based on semi-supervised discriminant analysis. Compared with the prior art, the beneficial effects of the financial distress prediction device based on semi-supervised discriminant analysis provided in this application are the same as the beneficial effects of the financial distress prediction method based on semi-supervised discriminant analysis provided in the above embodiments, and other technical features in the financial distress prediction device based on semi-supervised discriminant analysis are the same as the features disclosed in the methods of the above embodiments, and will not be repeated here.

[0213] It should be understood that the above are merely illustrative examples and do not constitute any limitation on the technical solutions of the present invention. In specific applications, those skilled in the art can make settings as needed, and the present invention does not impose any restrictions on this.

[0214] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this invention. In practical applications, those skilled in the art can select some or all of the workflow to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.

[0215] In addition, for technical details not described in detail in this embodiment, please refer to the financial distress prediction method based on semi-supervised discriminant analysis provided in any embodiment of the present invention, which will not be repeated here.

[0216] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0217] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0218] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory / random access memory, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0219] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A financial distress prediction method based on semi-supervised discriminant analysis, characterized in that, The financial distress prediction method based on semi-supervised discriminant analysis includes: Obtain a financial ratio dataset, which includes a training dataset, comprising a labeled dataset and an unlabeled dataset, wherein the labeled dataset includes labeled samples of a first category and labeled samples of a second category. The unlabeled dataset is input into the attribution discrimination model, which outputs the discrimination result. Based on the discrimination result, the unlabeled dataset is assigned an initial pseudo-label. The unlabeled dataset after being assigned the initial pseudo-label includes a first category pseudo-label sample and a second category pseudo-label sample. The attribution discrimination model is a DEA-DA model built based on multiple scale rewards. The attribution discrimination model is trained on the labeled dataset. The scale rewards include unlimited scale rewards, increasing scale rewards, decreasing scale rewards, and constant scale rewards. The TDEA-DA model is trained on the labeled dataset and the unlabeled dataset after being given initial pseudo-labels. The TDEA-DA model includes base classifiers corresponding to various scale rewards. The TDEA-DA model is a model obtained by improving and optimizing the DEA-DA model based on a semi-supervised learning framework. The DEA-DA model is a model constructed by integrating data envelopment analysis and discriminant analysis frameworks. In response to the training TDEA-DA model not satisfying the commutation condition, the target function values ​​corresponding to each base classifier are output. The TDEA-DA model is adjusted based on the objective function value to obtain the target prediction model, and financial distress prediction is performed on the financial sample to be predicted based on the target prediction model.

2. The financial distress prediction method based on semi-supervised discriminant analysis as described in claim 1, characterized in that, The training of the TDEA-DA model based on the labeled dataset and the unlabeled dataset after assigning initial pseudo-labels includes: Initialize the training parameters, which include the sample class weights and model parameters of the training dataset of the TDEA-DA model; The initialized training parameters, the labeled dataset, and the unlabeled dataset with initial pseudo-labels are input into the TDEA-DA model, and the first stage optimization of the TDEA-DA model is performed, outputting the model parameters and result sample set after the first stage optimization. Determine whether the resulting sample set is linearly separable; If the result sample set is linearly inseparable, then the second stage optimization of the TDEA-DA model is performed based on the result sample set and the model parameters optimized in the first stage, and the model parameters optimized in the second stage are output. The first category attribution strategy is determined based on the model parameters optimized in the second stage. Based on the first category attribution discrimination strategy, the samples in the training dataset of the TDEA-DA model are classified to obtain the trained TDEA-DA model; Determine whether the trained TDEA-DA model satisfies the commutation condition; If the trained TDEA-DA model satisfies the exchange condition, then the pseudo-labels of the pseudo-labeled sample pairs that satisfy the exchange condition are swapped, and the process returns to the first stage of optimization of the TDEA-DA model. If the trained TDEA-DA model does not meet the exchange condition, then the objective function values ​​corresponding to each base classifier are output, and the TDEA-DA model is adjusted based on the objective function values ​​to obtain the target prediction model.

3. The financial distress prediction method based on semi-supervised discriminant analysis as described in claim 2, characterized in that, After determining whether the result sample set is linearly separable, the method further includes: If the result sample set is linearly separable, then the second category classification strategy is determined based on the model parameters optimized in the first stage. Based on the second category classification strategy, each sample in the training dataset of the TDEA-DA model is classified to obtain the trained TDEA-DA model, and the step of determining whether the trained TDEA-DA model satisfies the exchange condition is executed.

4. The financial distress prediction method based on semi-supervised discriminant analysis as described in claim 3, characterized in that, The optimized model parameters in the first stage include the optimized weights of input features, the weights of output features, and the classification threshold parameter; the optimized model parameters in the second stage include the optimized classification boundary parameter, the weights of input features, and the weights of output features in the linearly inseparable scenario. The first category attribution strategy includes: in, This represents the test samples in the test set. Indicates test sample The Middle One input characteristic, Indicates test sample The Middle Individual output characteristics This represents the total number of input-type characteristics. This represents the total number of output-type characteristics. This represents the weight of the input features output after the second stage of optimization. This represents the weight of the output feature after the second stage of optimization. This represents the classification boundary parameters output after the second stage of optimization. This indicates a labeled dataset. D The first category contains labeled samples. This indicates a labeled dataset. D The second category contains labeled samples; The second category attribution strategy includes: in, This represents the weights of the input features output after the first stage of optimization. This represents the weight of the output features after the first stage of optimization. This represents the classification threshold parameter output after the first stage of optimization. This represents the interval of the discriminant function.

5. The financial distress prediction method based on semi-supervised discriminant analysis as described in claim 4, characterized in that, The exchange conditions include: Does the training dataset of the TDEA-DA model contain a pair of unlabeled samples with different pseudo-labels that satisfy the following condition: in, This represents the first category of pseudo-labeled samples. Positive slack variables, For the second category of pseudo-label samples Negative slack variables, This represents an unlabeled dataset. This represents the weight of the pseudo-labeled sample in the first category. This represents the weight of the pseudo-labeled sample in the second category.

6. The financial distress prediction method based on semi-supervised discriminant analysis as described in claim 5, characterized in that, The first phase of optimization includes: Model parameters are optimized using a first objective function and a first constraint condition, wherein the first objective function includes: in, This indicates a labeled dataset. D The first category contains labeled samples. This indicates a labeled dataset. D The second category contains labeled samples. Represents unlabeled datasets U The first category of pseudo-labeled samples, Represents unlabeled datasets U The second category of pseudo-label samples, This represents the positive slack variable for labeled samples in the first category. This represents the negative slack variable for labeled samples in the second category. Indicates the sample class weights. This represents the weight of the pseudo-labeled sample in the first category. The positive relaxation variable represents the pseudo-labeled samples of the first category. This represents the weight of the pseudo-labeled samples in the second category. The negative slack variable represents the pseudo-labeled samples of the second category; The first constraint includes: in, Indicates input-type characteristics, To input feature index, For sample index, Indicates input characteristics The weight, Indicates output-type characteristics, Indicates the output feature index. Indicates output characteristics The weight, This represents the classification threshold parameter. This represents a positive tolerance parameter, which provides tolerance space for constraint differences among labeled samples. , and Constraint variables representing returns to scale. This represents the negative slack variable for labeled samples in the first category. This represents the positive relaxation variable for labeled samples in the second category. This represents the negative slack variable for the first category of pseudo-labeled samples. The positive relaxation variable represents the pseudo-labeled samples of the second category. It is a positive real number; The result sample set includes: in, Represents the result sample set, , , and They represent in , , and The set of samples that were not correctly assigned to a specific category. This indicates that the discriminant function generated by the model classifies it as belonging to... There are categorized labeled samples. This indicates that the discriminant function generated by the model classifies it as belonging to... There are categorized labeled samples. This indicates that the discriminant function generated by the model classifies it as belonging to... Unlabeled samples of each category This indicates that the discriminant function generated by the model classifies it as belonging to... Unlabeled samples of each category This represents the weights of the input features output after the first stage of optimization. This represents the weight of the output features after the first stage of optimization. This represents the classification threshold parameter output after the first stage of optimization; The second phase of optimization includes: The model parameters are optimized using a second objective function and a second constraint condition. The second objective function includes: The second constraint includes: in, This represents the classification boundary parameters in linearly inseparable scenarios.

7. The financial distress prediction method based on semi-supervised discriminant analysis as described in any one of claims 1 to 6, characterized in that, The process of adjusting the TDEA-DA model based on the objective function value to obtain the target prediction model, and then using the target prediction model to predict financial distress for the financial sample to be predicted, includes: Based on the objective function value, the target weight values ​​of each base classifier in the TDEA-DA model are determined and adjusted, referring to the following formula: in, Indicates the first The target weight values ​​of each base classifier Indicates the first The target function value output by each base classifier This represents the sum of the reciprocals of the objective function values ​​of all base classifiers; The TDEA-DA model is adjusted based on the target weight values ​​to obtain the target prediction model, and financial distress prediction is performed on the financial sample to be predicted based on the target prediction model.

8. A financial distress prediction device based on semi-supervised discriminant analysis, characterized in that, The financial distress prediction device based on semi-supervised discriminant analysis includes: The dataset acquisition module is used to acquire a financial ratio dataset, which includes a training dataset, a labeled dataset and an unlabeled dataset, and the labeled dataset includes labeled samples of a first category and labeled samples of a second category. The pseudo-label assignment module is used to input the unlabeled dataset into the attribution discrimination model, output the discrimination result, and assign initial pseudo-labels to the unlabeled dataset based on the discrimination result. The unlabeled dataset after being assigned initial pseudo-labels includes first-class pseudo-label samples and second-class pseudo-label samples. The attribution discrimination model is a DEA-DA model built based on multiple scale rewards. The attribution discrimination model is trained based on the labeled dataset. The scale rewards include unlimited scale rewards, increasing scale rewards, decreasing scale rewards, and constant scale rewards. The model training module is used to train the TDEA-DA model based on the labeled dataset and the unlabeled dataset after assigning initial pseudo-labels. The TDEA-DA model includes base classifiers corresponding to various scale rewards. The TDEA-DA model is a model obtained by improving and optimizing the DEA-DA model based on a semi-supervised learning framework. The DEA-DA model is a model constructed by integrating data envelopment analysis and discriminant analysis. The output response module is used to output the objective function values ​​of each base classifier in response to the trained TDEA-DA model not satisfying the commutation condition. The financial distress prediction module is used to adjust the TDEA-DA model based on the objective function value to obtain the target prediction model, and to predict the financial distress of the financial sample to be predicted based on the target prediction model.

9. A financial distress prediction device based on semi-supervised discriminant analysis, characterized in that, The financial distress prediction device based on semi-supervised discriminant analysis includes: a memory, a processor, and a financial distress prediction program based on semi-supervised discriminant analysis stored in the memory. The processor is used to run the financial distress prediction program based on semi-supervised discriminant analysis, and the financial distress prediction program based on semi-supervised discriminant analysis is configured to implement the financial distress prediction method based on semi-supervised discriminant analysis as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a financial distress prediction program based on semi-supervised discriminant analysis, which, when executed by a processor, implements the financial distress prediction method based on semi-supervised discriminant analysis as described in any one of claims 1 to 7.

Images