A method and system for classifying imbalanced data based on a minority class sample confidence linkage mechanism.

Abstract

This invention discloses an imbalanced data classification method and system based on a minority class sample confidence linkage mechanism, belonging to the field of data mining. The invention includes: S1: acquiring the imbalanced dataset to be classified and preprocessing it to identify minority and majority class samples; S2: constructing confidence scores for minority class samples; S3: performing linked oversampling based on the confidence scores to generate synthetic minority class samples; S4: constructing triplet samples based on the confidence scores and setting an adaptive interval; S5: using a feature extraction model to learn embedding features on the triplet samples; S6: constructing majority class prototypes and minority class weighted prototypes in the embedding space and completing the classification. This invention uses the minority class sample confidence as a unified linkage variable throughout the entire process of sample generation, sample pairing, constraint adjustment, and classification decision-making, which can improve the identification ability, classification stability, and model generalization performance of minority class samples.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and data mining technology, and in particular relates to an imbalanced data classification method and system based on a minority class sample confidence linkage mechanism. Background Technology

[0002] Imbalanced data classification is a widespread problem in applications such as medical diagnosis, industrial fault detection, financial fraud detection, cybersecurity early warning, and bioinformatics analysis. In these tasks, minority class samples typically correspond to abnormal events, fault states, high-risk objects, or key pathological categories, and have higher identification value and application significance compared to majority class samples. Because the number of minority class samples is significantly less than that of majority class samples, traditional classification models are easily dominated by majority class samples during training, leading to insufficient ability to identify minority class samples.

[0003] Existing methods for classifying imbalanced data mainly include data-level sample augmentation methods and algorithm-level classification model optimization methods. Data-level methods typically expand minority class samples through duplication, interpolation, or generation to alleviate class imbalance; algorithm-level methods improve the model's ability to distinguish minority class samples by improving the loss function, optimizing the classification model structure, or introducing metric learning mechanisms. However, most existing technologies treat minority class sample augmentation, sample pairing, embedding learning, and classification as independent processing stages, making it difficult to continuously transfer and effectively utilize minority class sample quality information across multiple stages.

[0004] Furthermore, traditional methods typically lack a unified variable to characterize the reliability of minority class samples, and also lack a linkage mechanism to simultaneously apply this reliability to the entire process of sample generation, training constraints, and classification decisions. Therefore, in scenarios where minority class samples have complex boundaries, sparse local distributions, or significant differences in sample quality, problems such as noisy sample amplification, training constraint imbalance, and class center shift can still easily occur, thus limiting the final recognition performance of minority class samples.

[0005] Therefore, it is necessary to propose an imbalanced data classification method and system that takes the confidence level of minority class samples as the core and applies it in conjunction with the entire process of minority class sample generation, sample pairing, adaptive interval adjustment and minority class prototype construction. Summary of the Invention

[0006] In view of this, the present invention provides an imbalanced data classification method and system based on a minority class sample confidence linkage mechanism. By constructing a minority class sample confidence score and using the minority class sample confidence score as a unified linkage variable, it simultaneously applies to minority class sample generation, triple sample construction, adaptive interval setting, and minority class weighted prototype construction, thereby achieving efficient classification of imbalanced data. This solves the technical problems in the prior art where the minority class sample enhancement process and classification discrimination process are separated, sample quality information is difficult to continuously transmit, and the minority class sample recognition ability is insufficient.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] An imbalanced data classification method based on a minority class sample confidence linkage mechanism includes the following steps:

[0009] S1: Obtain the imbalanced dataset to be classified, and preprocess the imbalanced dataset to identify minority class samples and majority class samples to obtain preprocessed training data.

[0010] S2: Construct confidence scores for minority class samples;

[0011] S3: Using the confidence score of the minority class samples as the control variable in the minority class sample generation process, perform linked oversampling to generate synthetic minority class samples;

[0012] S4: Using the confidence scores of the minority class samples as the basis for sample pairing and constraint strength adjustment, construct triplet samples and set an adaptive interval;

[0013] S5: Use a feature extraction model to learn the embedding features of the triplet samples to obtain the sample embedding representation;

[0014] S6: Construct majority class prototypes and minority class weighted prototypes in the embedding space, and complete the classification and discrimination based on the distance between the test sample and the prototypes.

[0015] Furthermore, in step S2, the confidence score of the minority class sample is jointly determined by local density, boundary security, and distribution offset, and is used to comprehensively evaluate the quality of the minority class sample from multiple aspects such as local structural stability, boundary position security, and global distribution representativeness.

[0016] Furthermore, in step S3, the linked oversampling uses the minority class sample confidence score as the basis for seed sample selection, interpolation object selection, and interpolation amplitude control, so that the minority class sample quality information directly affects the sample generation process.

[0017] Furthermore, in step S4, the triplet sample construction uses the minority class sample confidence score for anchor sample selection, positive and negative sample pairing, and adaptive interval setting, so that sample quality information continues to play a role in the metric learning stage.

[0018] Furthermore, in step S5, the feature extraction model is a multi-branch feature extraction model with shared weights. The model performs a unified embedding mapping on anchor samples, positive samples, and negative samples, and is trained using a metric loss corresponding to the adaptive interval.

[0019] Furthermore, in step S6, the minority class weighted prototype is constructed by weighting the embedding representation of minority class samples with confidence, so as to improve the contribution of highly reliable minority class samples in classification decision.

[0020] This invention also provides an imbalanced data classification system based on a minority class sample confidence linkage mechanism to implement the above method, comprising a data acquisition module, a data preprocessing module, a confidence assessment module, a linkage oversampling module, a linkage triplet construction module, a linkage feature learning module, and a linkage classification and discrimination module connected in sequence.

[0021] As can be seen from the above technical solution, compared with the prior art, the present invention has at least the following beneficial effects:

[0022] 1) This invention uses the confidence level of minority class samples as a unified linkage variable, and integrates the quality information of minority class samples throughout the entire process of sample generation, sample pairing, training constraints and classification, thus avoiding the problem of the separation of each stage in the prior art.

[0023] 2) This invention introduces a confidence control mechanism in the minority class sample generation process, which can reduce the probability of low-quality minority class samples being amplified without constraints and improve the rationality of the distribution of synthesized minority class samples;

[0024] 3) This invention introduces a confidence linkage mechanism in the construction of triples and the setting of intervals, so that minority class samples of different qualities can obtain differentiated constraints and improve the inter-class separability in the embedding space;

[0025] 4) This invention introduces a confidence weighting mechanism in the minority class weighted prototype construction process, so that highly confident minority class samples contribute more to the classification decision, thereby improving the minority class sample recognition ability, classification stability and model generalization performance. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0027] Figure 1 A flowchart of an imbalanced data classification method based on a minority class sample confidence linkage mechanism provided in an embodiment of the present invention;

[0028] Figure 2 A schematic diagram illustrating the principle framework of an imbalanced data classification method based on a minority class sample confidence linkage mechanism, provided in an embodiment of the present invention.

[0029] Figure 3 This is a structural block diagram of an imbalanced data classification system based on a minority class sample confidence linkage mechanism, provided for an embodiment of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus 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 apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0032] See Figure 1 and Figure 2 This invention provides an imbalanced data classification method based on a minority class sample confidence score linkage mechanism. This method uses the minority class sample confidence score as a unified linkage variable, applying it throughout the entire process of data augmentation, sample pairing, constraint adjustment, and classification, thereby improving the identification ability of minority class samples and the overall classification performance. The method includes the following steps.

[0033] S1: Obtain the imbalanced dataset and preprocess it.

[0034] An imbalanced dataset containing majority and minority class samples is obtained, and the imbalanced dataset is preprocessed to obtain the training data required for subsequent modeling.

[0035] The preprocessing includes: standardizing the sample features to eliminate dimensional differences between different features; deleting features with a missing value ratio exceeding a preset threshold to improve data quality; filling the remaining missing values ​​with imputation; and filtering features when the feature dimension of the dataset exceeds a preset upper limit to reduce computational complexity.

[0036] In one implementation, the sample features are standardized with zero mean and unit variance; features with a missing value ratio exceeding 25% are deleted; the remaining missing values ​​are filled using the K-nearest neighbor imputation method; when the feature dimension exceeds 3000, the variance screening method is used to retain the top 3000 features with the largest variance.

[0037] In this embodiment, let the original sample feature matrix be... The total number of samples is The feature dimension is Among them, matrix Each row corresponds to a sample, and each column corresponds to a feature dimension. By standardizing the original features, the impact of differences in feature dimensions on subsequent distance calculations and model training can be reduced.

[0038] After the above preprocessing, majority and minority class samples are identified based on the number of samples in each class. The class with more samples is designated as the majority class, and the class with fewer samples is designated as the minority class. This yields the preprocessed training data required for constructing the confidence scores of the minority class samples and performing linked classification.

[0039] S2: Construct confidence scores for minority class samples

[0040] The confidence score of the minority class samples is not determined based on a single indicator, but is jointly constructed by local density, boundary security, and distribution shift, so as to comprehensively evaluate the sample quality from three aspects: local structural stability, boundary position security, and global distribution representativeness of the minority class samples.

[0041] In the implementation method, for any minority class sample Its local density It can be determined by the distance between the sample and its minority class nearest neighbors, and its expression is:

[0042]

[0043] in, Indicates the first A minority class of samples, Indicates sample The A minority of nearest neighbor samples, Describing the Euclidean norm, Indicates the number of nearest neighbors. This indicates that the summation is performed on the selected nearest neighbor samples. This represents the smoothing constant.

[0044] Its boundary security It can be determined by the average distance of the sample to the majority class nearest neighbors and the average distance to the minority class nearest neighbors, and its expression is:

[0045]

[0046] in, Indicates sample The average distance to the nearest neighbor samples of the majority class. Indicates sample The average distance to the nearest neighbor samples of the minority class This represents the smoothing constant.

[0047] Its distribution shift It can be determined by the distance from the sample to the center of the minority class samples, and its expression is:

[0048]

[0049] in, Indicates the minority class sample centers. This represents the maximum distance among all minority class samples to the minority class center. This represents the smoothing constant.

[0050] After normalizing the above three quantities, the minority class sample confidence score It can be represented as:

[0051]

[0052] in, Indicates the first The confidence scores corresponding to the minority class samples. Represents the reciprocal of local density. This represents the normalized result of the inverse of the local density. The normalized result representing boundary security. This represents the normalized result of the distribution offset complement. , , These represent the weight coefficients corresponding to the three components.

[0053] S3: Perform linked oversampling based on minority class sample confidence.

[0054] The above method allows us to obtain the quality characterization results of each minority class sample in the subsequent linked classification process. A higher confidence score indicates greater reliability of the minority class sample in terms of local structure, boundary location, and distribution representativeness. The key to the linked oversampling is using the minority class sample confidence score as a unified control variable in the minority class sample generation process. Specifically, the minority class sample confidence score is used as the seed sample sampling probability, prioritizing high-confidence minority class samples as interpolation seeds; among the nearest neighbors of the interpolation seed, samples with similar confidence scores are prioritized as interpolation objects; and the interpolation coefficients are determined based on the minority class sample confidence score of the interpolation seed, thereby controlling the offset of the generated samples relative to the original minority class samples.

[0055] In one embodiment, the interpolation coefficients Determine using the following formula:

[0056]

[0057] in, Indicates the first Interpolation coefficients corresponding to each interpolation seed sample Indicates the lower limit of the interpolation coefficients. This indicates the upper limit of the interpolation coefficients. This represents the confidence score of the minority class sample corresponding to the interpolation seed sample.

[0058] Synthetic minority class samples can be generated using the following formula:

[0059]

[0060] in, This represents the generated synthetic minority class samples. Indicates the interpolation seed sample. Represents the interpolation object sample. This represents the feature difference vector between the interpolation object sample and the interpolation seed sample. This represents the disturbance noise term.

[0061] In this way, the confidence score of the minority class samples not only determines which minority class samples are more suitable as the basis for amplification, but also determines the offset strength of the synthesized sample relative to the original sample, thereby making the generated sample more consistent with the minority class distribution structure.

[0062] S4: Construct linked triples based on minority class sample confidence.

[0063] Triple samples consisting of anchor samples, positive samples, and negative samples are constructed, and the sample pairing and constraint strength are adjusted in conjunction with the minority class sample confidence score. Specifically, minority class samples are preferentially used as anchor samples; positive samples are selected based on the feature distance and confidence difference between the anchor sample and candidate samples of the same class; negative samples are selected based on the feature distance between the anchor sample and candidate samples of different classes; and the adaptive interval corresponding to the triple samples is set according to the minority class sample confidence score of the anchor sample.

[0064] In one embodiment, the adaptive interval It can be represented as:

[0065]

[0066] in, Indicates the first The adaptive interval corresponding to each triplet sample. Indicates the basic interval. This represents the interval adjustment coefficient. This represents the confidence score of the minority class of the anchor sample.

[0067] In this way, the confidence score of the minority class samples continues to affect the sample pairing and interval adjustment process, so that minority class samples of different qualities receive training constraints of different intensities during the metric learning stage, thereby improving the inter-class separability in the embedding space.

[0068] S5: Perform linked feature learning

[0069] The triplet samples are input into the feature extraction model for embedding feature learning. This step is not limited to a specific network structure, but rather focuses on "the confidence of minority class samples participating in the adjustment of training constraints".

[0070] In one embodiment, the feature extraction model is a multi-branch feature extraction model with shared weights, which maps anchor samples, positive samples, and negative samples to a unified embedding space, and is trained using a metric loss corresponding to the adaptive interval to improve the separability of minority class samples and majority class samples in the embedding space.

[0071] In one implementation, the loss function used can be expressed as:

[0072]

[0073] in, Indicates training loss in triplet, This represents the total number of triplet samples. This represents the embedding mapping function corresponding to the feature extractor. Indicates the first Anchor point samples in a triplet Indicates the first Positive samples in a triplet Indicates the first Negative samples in a triplet Indicates the first The adaptive intervals corresponding to the triplet samples are obtained through this step. Sample embedding representations can be used for subsequent prototype construction and classification.

[0074] S6: Classification and discrimination based on minority class weighted prototypes

[0075] After obtaining the sample embedding representation, a majority class prototype and a minority class weighted prototype are constructed, and classification is performed based on the distance between the test sample and the prototype.

[0076] The majority class prototype is constructed using the mean of the majority class sample embeddings; the minority class weighted prototype is constructed using minority class sample embeddings weighted by the confidence scores of the minority class samples. This approach gives higher weight to high-confidence minority class samples in the minority class weighted prototype, thereby reducing the interference of low-confidence minority class samples on the class center.

[0077] In one implementation, the distances from the test sample to the minority class weighted prototype and the majority class prototype are denoted as follows: and Its expression is as follows:

[0078]

[0079] in, Indicates the sample to be tested. This represents the representation of the sample to be tested in the embedding space. Represents the weighted prototype of the minority class. This represents the prototype of the majority class.

[0080] like If the probability of a sample belonging to the minority class is positive, it is determined that the sample belongs to the minority class; otherwise, it is determined that the sample belongs to the majority class. Furthermore, the predicted probability of a sample belonging to the minority class can be calculated using the following formula:

[0081]

[0082] in, Indicates the predicted probability of the minority class. This represents the smoothing constant.

[0083] Experimental results show that the minority class sample confidence linkage mechanism adopted in this embodiment can effectively improve the representation quality of minority class samples in imbalanced data and enhance the stability of the classification model in terms of accuracy, recall and comprehensive discrimination ability.

[0084] In one embodiment, to verify the effectiveness and applicability of the method of the present invention, 18 imbalanced datasets were selected for experimental verification, including 11 cancer metastasis datasets in the TCGA database and 7 popular datasets in the UCI machine learning database.

[0085] The TCGA database contains 11 cancer datasets covering three biometric data types: mRNA expression data, microRNA expression data, and DNA methylation data, corresponding to the M0 and M1 classes in a binary classification task. M0 represents samples without distant metastasis, while M1 represents samples with distant metastasis. Because the number of M1 samples is significantly less than that of M0 samples, an imbalanced data distribution is formed.

[0086] The seven popular datasets in the UCI Machine Learning Database cover multiple application areas such as medical diagnosis and physical detection, and are used to verify the applicability and generalization ability of the method of this invention in non-cancer imbalanced data classification tasks.

[0087] In one implementation, the 11 cancer datasets in the TCGA database are shown in Table 1, and the 7 popular datasets in the UCI machine learning database are shown in Table 2.

[0088] Table 1. Datasets for 11 types of cancer in the TCGA database

[0089]

[0090] Table 2 Seven popular datasets in the UCI database

[0091]

[0092] Corresponding to the above methods, this invention also provides an imbalanced data classification system based on a minority class sample confidence linkage mechanism, the structural block diagram of which is shown below. Figure 3 As shown, it includes a data acquisition module, a data preprocessing module, a confidence assessment module, a linked oversampling module, a linked triplet construction module, a linked feature learning module, and a linked classification and discrimination module.

[0093] The system comprises the following modules: a data acquisition module for acquiring an imbalanced dataset to be classified; a data preprocessing module for preprocessing the imbalanced dataset and identifying minority and majority class samples; a confidence assessment module for constructing confidence scores for minority class samples; a linked oversampling module for generating synthetic minority class samples based on the minority class sample confidence scores; a linked triplet construction module for constructing triplet samples and setting adaptive intervals; a linked feature learning module for training a feature extraction model and outputting sample embedding representations; and a linked classification and discrimination module for constructing majority class prototypes and minority class weighted prototypes, and outputting classification results and predicted probabilities.

[0094] For the system or system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and relevant details can be found in the description of the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0095] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for classifying imbalanced data based on a minority class sample confidence linkage mechanism, characterized in that, Includes the following steps: S1: Obtain the imbalanced dataset to be classified, and preprocess the imbalanced dataset to identify minority class samples and majority class samples to obtain preprocessed training data. S2: Construct confidence scores for minority class samples; S3: Using the confidence scores of the minority class samples as control variables in the minority class sample generation process, perform linked oversampling to generate synthetic minority class samples and construct an augmented training dataset; S4: Using the confidence scores of the minority class samples as the basis for sample pairing and constraint strength adjustment, construct triplet samples consisting of anchor samples, positive samples, and negative samples, and set an adaptive interval for each triplet sample. S5: Input the triplet samples into the feature extraction model for embedding feature learning, and train the feature extraction model using a metric constraint associated with the confidence score of the minority class samples to obtain the sample embedding representation; S6: Construct majority class prototypes and minority class weighted prototypes in the sample embedding space, and complete classification prediction based on the distance between the sample to be tested and the majority class prototypes and the minority class weighted prototypes.

2. The imbalanced data classification method according to claim 1, characterized in that, In step S2, the confidence score of the minority class sample is jointly determined by local density, boundary safety, and distribution offset; wherein, local density is used to characterize the density of the distribution of minority class samples in the local neighborhood of the minority class, boundary safety is used to characterize the positional safety of the minority class sample relative to the boundary of the majority class sample, and distribution offset is used to characterize the degree of deviation of the minority class sample relative to the center of the minority class sample.

3. The imbalanced data classification method according to claim 2, characterized in that, In step S2, minority class samples Local density Boundary security and distribution offset Calculate them as follows: After normalizing the local density, boundary safety, and distribution shift, the confidence scores for minority class samples are constructed according to the following formula. : in, Indicates the first A minority class of samples, Indicates sample The A minority of nearest neighbor samples, Describing the Euclidean norm, Indicates the number of nearest neighbors. Represents the smoothing constant; Indicates sample The average distance to the nearest neighbor samples of the majority class. Indicates sample The average distance to the nearest neighbor samples of the minority class; Indicates the center of the minority class samples; Indicates the first The confidence scores corresponding to each minority class sample; , , These represent the weight coefficients corresponding to the inverse of the local density term, the boundary safety term, and the distribution offset complement term, respectively.

4. The imbalanced data classification method according to claim 1, characterized in that, The linked oversampling in step S3 includes: using the minority class sample confidence score as the seed sample sampling probability, selecting a minority class sample as the interpolation seed; selecting samples with confidence scores close to the interpolation seed from among the nearest neighbors of the same class as the interpolation seed as the interpolation object; and determining the interpolation coefficients based on the minority class sample confidence score of the interpolation seed. The interpolation coefficients are calculated according to the following formula: And generate synthetic minority class samples according to the following formula: in, Indicates the first Interpolation coefficients corresponding to each interpolation seed sample Indicates the lower limit of the interpolation coefficients. This indicates the upper limit of the interpolation coefficients. This represents the confidence score of the minority class sample corresponding to the interpolation seed sample; This represents the generated synthetic minority class samples. Indicates the interpolation seed sample. Represents the interpolation object sample. This represents the disturbance noise term.

5. The imbalanced data classification method according to claim 1, characterized in that, The construction of the triplet samples in step S4 includes: prioritizing minority class samples as anchor samples; selecting positive samples based on the feature distance and confidence difference between the anchor samples and candidate samples of the same class; selecting negative samples based on the feature distance between the anchor samples and candidate samples of different classes; and setting an adaptive interval corresponding to the triplet samples based on the minority class confidence score of the anchor samples. Calculate using the following formula: in, Indicates the first The adaptive interval corresponding to each triplet sample. Indicates the basic interval. This represents the interval adjustment coefficient. This represents the confidence score of the minority class of the anchor sample.

6. The imbalanced data classification method according to claim 1, characterized in that, The feature extraction model described in step S5 is a multi-branch feature extraction model with shared weights. This model performs a unified embedding mapping on anchor samples, positive samples, and negative samples, and is trained using the following loss function: in, Indicates training loss in triplet, Indicates the number of triplet samples. This represents the embedding mapping function corresponding to the feature extractor. Indicates the first Anchor point samples in a triplet Indicates the first Positive samples in a triplet Indicates the first Negative samples in a triplet Indicates the first The adaptive interval corresponding to each triplet sample.

7. The imbalanced data classification method according to claim 1, characterized in that, In step S6, the majority class prototypes The prototype is constructed using the mean of the majority class sample embeddings and the minority class weighted prototype. The minority class embedding representation is constructed using a weighted minority class sample confidence score; the distances from the test sample to the minority class weighted prototype and the majority class prototype are denoted as follows: and Its expression is as follows: in, Indicates the sample to be tested. This represents the representation of the sample to be tested in the embedding space. Represents the weighted prototype of the minority class. Represents the prototype of the majority class; if If the result is positive, the sample is determined to belong to the minority class; otherwise, it is determined to belong to the majority class. The predicted probability of the minority class is then obtained according to the following formula. : in, This represents the predicted probability that the sample to be tested belongs to the minority class. This represents the smoothing constant.

8. An imbalanced data classification system based on a minority class sample confidence linkage mechanism, characterized in that, The method for implementing the imbalanced data classification method as described in any one of claims 1-7 includes, in sequence, a data acquisition module, a data preprocessing module, a confidence assessment module, a linked oversampling module, a linked triplet construction module, a linked feature learning module, and a linked classification and discrimination module; The data acquisition module is used to acquire the imbalanced dataset to be classified; The data preprocessing module is used to preprocess the imbalanced dataset and automatically identify minority and majority class samples. The confidence assessment module is used to calculate the confidence score for minority class samples; The linked oversampling module is used to generate synthetic minority class samples based on the confidence scores of minority class samples; The linked triplet construction module is used to construct triplet samples with adaptive intervals based on the confidence scores of minority class samples. The linked feature learning module is used to train a shared weight feature extraction model and output sample embedding representations; The linked classification and discrimination module is used to construct majority class prototypes and minority class weighted prototypes, and output classification results and predicted probabilities.

Images