Multi-density data classification method and system based on density hierarchical clustering

Abstract

The invention discloses a multi-density data classification method and a multi-density data classification system based on density hierarchical clustering, and belongs to the technical field of multi-dimensional data classification. According to the method, a Gaussian mixture model is used for carrying out density stratification on various types of data, data in different density layers are separated, then a DBSCAN algorithm is used for identifying various data areas in each density layer, and the classification accuracy is improved. Forming a subcategory; and finally, identifying and summarizing subclasses by using a Bayesian model, thereby realizing classification of multi-density data. According to the method, the internal structure of the data can be described more accurately, so that the Bayesian classifier can adapt to the characteristics of regions with different densities more accurately.

Description

Technical Field

[0001] The present invention belongs to the technical field of multidimensional data classification with multi-density characteristics, and in particular relates to a multi-density data classification method and system based on density hierarchical clustering. Background Art

[0002] Traditional clustering algorithms, such as spectral clustering and DBSCAN, are not suitable for multi-density data. Although traditional density-based algorithms can help process multi-density data, they usually require a large number of parameters to be set, which increases the complexity of model application and inevitably brings uncertainty in classification. For example, when monitoring water bodies, sensors usually collect multiple physical and biochemical indicators at the same time, thus forming data with multi-dimensional characteristics. When processing these data, we often face complex and diverse density distributions of data. If water quality abnormalities occur less frequently, then the data involved in this classification in the collected data will naturally be less. From the final collected data, there are data with multi-density characteristics in the data space. For another example, in the field of meteorology, some weather phenomena occur less frequently. In order to accurately distinguish the less frequent weather phenomena from the data with very different probabilities of occurrence, data classification with multi-density characteristics is also involved. However, this kind of data brings great challenges to the classification of data using Bayesian classifiers. Therefore, we need a method to effectively process multi-density distribution data, improve the classification performance of Bayesian classifiers on multi-density data, and achieve accurate classification of unknown data. Summary of the invention

[0003] In order to solve the problems existing in the prior art, the present invention provides a multi-density data classification method and system based on density hierarchical clustering. For the classification problem of multi-density data, density stratification is used to generate subclasses, and a Bayesian classifier is used to achieve accurate classification of multi-density data.

[0004] In order to achieve the above-mentioned invention object, the present invention provides the following technical solutions:

[0005] The present invention provides a multi-density data classification method based on density hierarchical clustering, which is characterized by comprising the following steps:

[0006] Obtain a data set with category labels, divide the original data using the original category labels, and divide the data with the same category labels into the same data set, thereby obtaining multiple single-category data sets and corresponding category labels;

[0007] The obtained single-category data sets are respectively subjected to label prediction and clustering using density stratification algorithms to obtain corresponding density stratified data and corresponding density stratified labels;

[0008] Each data layer in the obtained density stratified data is clustered using the density clustering algorithm DBSCAN to generate new category labels;

[0009] Use the new category label and Bayesian formula to fit the Bayesian model for each single-category data set to achieve data classification;

[0010] The fitted Bayesian model is used to calculate the probability that the data point to be predicted belongs to the new category in each single-category data set, and then the probability that the data point to be predicted belongs to the original category is calculated;

[0011] Perform category probability calculation for the data point to be predicted, select the category with the largest probability as its predicted classification label, and return the predicted category label.

[0012] Furthermore, using the density stratification algorithm for a single-category dataset includes the following steps:

[0013] Through the density formula Calculate the single-category dataset U m For each data point x to be predicted i The data density is obtained by m , where K i is the data point x to be predicted in the single-category data set i The set of k data points closest to each other, E(x i ,x j ) represents the data point x i With data point x j The Euclidean distance of

[0014] For density data D m Perform EM fitting of a mixed Gaussian distribution with model number k to calculate ICL. This operation is performed iteratively between 2 and T, and the k with the maximum ICL is selected. m As the number of mixed Gaussian models, where T is the maximum number of models;

[0015] Select the optimal number of models k m Then, the density data D is analyzed using the mixed Gaussian model. m Make predictions and get new density stratified labels According to the model fitting results, the original data is divided into different density layers to obtain density layered data Layers m .

[0016] Furthermore, the data density is for each data point x i The reciprocal of the average distance between a sample and the k closest data points.

[0017] Furthermore, the mixed Gaussian distribution EM fitting EM refers to the EM algorithm, also known as the expectation maximization algorithm, the output of EM is a specified number of Gaussian model parameters {Θ1, Θ2, ...}, ICL refers to the integrated complete likelihood, and the calculation formula of ICL is as follows:

[0018]

[0019] in, is the degree of fit of the model, V m,K is the number of free parameters in the mixture model m containing K distributions, n m For a single-category dataset U m The number of data points to be predicted in .

[0020] Furthermore, the mixed Gaussian model is expressed as:

[0021]

[0022] Among them, H is the number of data density layers, Θ h is the parameter of the h-th Gaussian model, α h is the weight of the i-th Gaussian model, N(ρθ h ) is the probability density function of the h-th Gaussian model,

[0023] Furthermore, the density clustering algorithm DBSCAN uses two parameters, namely, the neighborhood radius ε and the minimum number of neighbors MinPts, wherein the square root of the number of data points is generally selected as the k value of DBSCAN, and the average distance k of the density layer is used as ε.

[0024] Furthermore, the Bayesian formula used to fit the Bayesian model is:

[0025]

[0026] Among them, P(C m a) is to calculate the new input feature vector α in category C m The concept of posterior under P(C m ) is a priori concept, P(C m A i ) is feature A i In Category C m The kernel probability density function under i,j C m ) is the conditional probability density function of Gaussian naive Bayes, μ j and Category C m The mean and variance of the jth feature, αi,j is the value of the jth feature of the i-th sample; P(a i |C m ) is the decision function of the multivariate Gaussian Bayes, μ m and Σ m C m The mean vector and covariance matrix of the class attributes, α i is the feature vector of the i-th sample, and p is the number of features.

[0027] Furthermore, the data point x to be predicted in each density layer data i The corresponding probability of the new category is expressed as:

[0028]

[0029] The data points x to be predicted in each density layer data i The corresponding probability of the original category is expressed as:

[0030]

[0031] Among them, x i For a single-category dataset U m The data points to be predicted in For x i Belongs to a new category The probability of For new categories The prior probability of For a given new category Under the condition of i The probability, P(x i ) is x i The marginal probability, P(C m x i ) is x i Belongs to the original category C m The probability of t' m For a single-category dataset U m The total number of clusters generated after DBSCAN clustering.

[0032] The present invention also provides a multi-density data classification system of density hierarchical clustering, comprising:

[0033] Data segmentation module: obtains labeled data sets, divides the original data using the original labels, and divides the data with the same labels into the same data set, thereby obtaining multiple single-category data sets and corresponding category labels;

[0034] Density stratification module: Use density stratification algorithm to perform label prediction and clustering on the obtained single-category data sets to obtain corresponding density stratification data and corresponding density stratification labels;

[0035] DBSCAN clustering module: Use the density clustering algorithm DBSCAN to cluster each data layer in the obtained density stratified data and generate new category labels;

[0036] Bayesian model fitting module: Use new category labels and Bayesian formula to fit the Bayesian model for each single-category data set to achieve data classification; Use the Bayesian formula to fit the Bayesian model to use different Bayesian classifiers based on different prior assumptions, such as Naive Bayes, Gaussian Naive Bayes, etc.;

[0037] Category probability calculation module: Use the fitted Bayesian model to calculate the data points to be predicted in each single-category data set, calculate the probability that the data points to be predicted belong to the new category, and calculate the probability that the data points to be predicted belong to the original category;

[0038] Category prediction module: performs category probability calculation for the data point to be predicted, selects the category with the largest probability as its predicted classification label, and returns the predicted category label.

[0039] Furthermore, the density stratification module includes:

[0040] Density data calculation module: calculates the density data of each single-category data set through the density formula;

[0041] Model fitting and selection module: Model fitting and optimal model number selection based on calculated density data;

[0042] Density layer prediction module: After selecting the optimal number of models, the Gaussian mixture model is used to predict the density data to obtain density layer labels; the original data points are divided into different density layers according to the model fitting results to obtain density layer data.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1. The classification method provided by the present invention can effectively cope with the complexity of data distribution when processing multi-density data. First, the Gaussian mixture model is used to divide the data into different density layers, and clustering is performed in different density layers after separation. The DBSCAN clustering algorithm is used to identify various data areas in each density layer to form subcategories, thereby more accurately describing the internal structure of the data. Finally, the fitted Bayesian model is used to realize the identification and aggregation of subcategories, thereby realizing the classification of multi-density data;

[0045] The Bayesian classification can more accurately adapt to the characteristics of different density areas, thereby realizing accurate classification of multi-density data; the present invention fully considers the local density information of the data in the classification process, and can better adapt to the data distribution characteristics of different density areas by performing local density estimation and hierarchical clustering on the data, thereby improving the adaptability and generalization ability of the classifier to the data;

[0046] 2. The method and classification system provided by the present invention have high flexibility and interpretability, and can flexibly select the parameters of density hierarchical clustering and the parameters of the Bayesian classifier according to actual conditions to meet the needs of different data sets and tasks. At the same time, since the present invention uses a Bayesian classifier for classification, it can provide interpretability of the classification results, and can clearly express the classification basis and reasoning process of the classifier, making the classification results more convincing and credible. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A comparison diagram of the Bayesian classification effects of the traditional Bayesian classification and the density hierarchical clustering of the present invention for 4 synthetic data sets;

[0048] Figure 2 It is a schematic diagram of the principle architecture of the multi-density data classification system based on density hierarchical clustering of the present invention;

[0049] Figure 3 It is a structural schematic diagram of the density stratification algorithm used by the density stratification module in the present invention. DETAILED DESCRIPTION

[0050] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way. On the contrary, these embodiments are provided to make the present disclosure more thorough and complete, and to fully convey the scope of the present disclosure to those skilled in the art.

[0051] According to an embodiment of the present invention, a multi-density data classification method and system based on density hierarchical clustering is proposed, which decomposes complex data into different density layers and then clusters them separately in different density layers, and finally aggregates the subclasses to form the final accurate classification.

[0052] Example 1

[0053] This embodiment provides a multi-density data classification method based on density hierarchical clustering, comprising the following steps:

[0054] S1. Data segmentation: Get the labeled data set DS = (U, Y), and use the original label Y = {y1, y2, ..., y n}For the original data U={x1,x2,...,x n} to divide the data with the same label into the same data set, thus obtaining multiple single-category data sets U1, U2, ..., U h And the corresponding category label C = {C1, C2, ..., C h}; where n is the number of samples, h is the number of labels, and there is no duplication of labels in C;

[0055] S2, density stratification: for the obtained single-category data sets U1, U2, ....., U h Use density stratification algorithms for label prediction and clustering respectively to obtain corresponding density stratification data Layers1, Layers2, ..., Layers h and the corresponding density stratification labels in, Medium m For a single-category dataset U m The number of clusters after clustering, 1≤m≤h, m∈Z;

[0056] The density stratification algorithm is as follows:

[0057] First, for the single-category dataset U m , through the density formula Calculate the single-category dataset U m For each data point x to be predicted i The data density is for each data point x i The reciprocal of the average distance between the sample and the k nearest data points is calculated by the formula Get density data D m , Among them, n m For U m The number of data points to be predicted in K i is the data point x to be predicted in the single-category data set i The set of k data points closest to each other, E(x i ,x j ) represents the data point x i With data point x j The Euclidean distance of

[0058] Secondly, for the density data D m Perform EM fitting of a mixed Gaussian distribution with model number k to calculate ICL. This operation is performed iteratively between 2 and T, and the k with the maximum ICL is selected. m As the number of mixed Gaussian models, where T is the maximum number of models;

[0059] The output of EM in the mixed Gaussian distribution EM fitting is a specified number of Gaussian model parameters {Θ1, Θ2, ...}, and ICL refers to the integrated complete likelihood. The calculation formula of ICL is as follows:

[0060]

[0061] in, is the degree of fit of the model, V m,K is the number of free parameters in the mixture model m containing K distributions, n m For a single-category dataset U m The number of data points to be predicted in ;

[0062] Finally, select the optimal number of models k m Then, the density data D is analyzed using the mixed Gaussian model. m Make predictions and get new density stratified labels According to the model fitting results, the original data is divided into different density layers to obtain density layered data Layers m ;

[0063] The Gaussian mixture model is expressed as:

[0064]

[0065] Among them, H is the number of data density layers, Θ h is the parameter of the h-th Gaussian model, α h is the weight of the i-th Gaussian model, N(ρθ h ) is the probability density function of the h-th Gaussian model,

[0066] S3, DBSCAN clustering: obtain density layered data Layers m Each data layer in is further clustered using the density clustering algorithm DBSCAN to generate new category labels in, t' m For a single-category dataset U m The total number of clusters generated after DBSCAN clustering, 1≤m≤h, m∈Z;

[0067] Among them, the density clustering algorithm DBSCAN uses two parameters, namely the neighborhood radius ε and the minimum number of neighbors MinPts. The square root of the number of data points is generally selected as the k value of DBSCAN, and the average distance k of the density layer is used as ε;

[0068] S4. Bayesian model fitting: For each single-category dataset Um Using new category labels As well as fitting the Bayesian model with the Bayesian formula, the Bayesian model is established using the characteristics of the data and prior knowledge, thereby realizing the classification of the data. Different Bayesian classifiers, such as Naive Bayes or Gaussian Naive Bayes, can be used according to different prior assumptions;

[0069] like Figure 1 As shown, it is a comparison diagram of the traditional Bayesian classification of 4 synthetic data sets and the Bayesian classification effect of density hierarchical clustering of the present invention; a is the True label row representing the true classification; b is GNB representing Gaussian naive Bayes, c is GNB-GDSC representing Gaussian naive Bayes of density hierarchical clustering, d is KNB representing kernel naive Bayes, and e is KNB-GDSC representing kernel naive Bayes of density hierarchical clustering. It can be seen from the figure that the density hierarchical clustering method based on the present embodiment can significantly improve the classification effect of the Bayesian classifier on multi-density data sets; taking the bth row as an example, we can see that in the traditional Gaussian naive Bayes method, due to its sensitivity to the data distribution assumption, the performance of the algorithm is greatly affected, and it cannot adapt well to multi-density data, resulting in a decrease in classification accuracy; in contrast, the Gaussian naive Bayes of Gaussian density hierarchical clustering shown in the cth row is not affected by the characteristics of multi-density data, and still maintains a high classification performance, and can better adapt to multi-density data structures, thereby improving the accuracy and robustness of classification.

[0070] In this embodiment, the Bayesian formula used to fit the Bayesian model is:

[0071]

[0072]

[0073] Among them, P(C m a) is to calculate the new input feature vector α in category C m The concept of posterior under P(C m ) is a priori concept, P(C m A i ) is feature A i In Category C m The kernel probability density function under i,j C m ) is the conditional probability density function of Gaussian naive Bayes, μ j and Category C m The mean and variance of the jth feature, α i,j is the value of the jth feature of the i-th sample; P(a i |C m) is the decision function of the multivariate Gaussian Bayes, μ m and Σ m C m The mean vector and covariance matrix of the class attributes, α i is the feature vector of the i-th sample, and p is the number of features;

[0074] S5. Category probability calculation: For each single category data set U m The data point x to be predicted in i Use the fitted Bayesian model to calculate the data point x to be predicted i Belongs to a new category Probability The original category C m Different new categories The probability of the data point to be predicted is summed up to get i Belongs to the original category C m The probability P(C m x i );

[0075] Among them, the data point x to be predicted in each density layer data i The corresponding probability of the new category is expressed as:

[0076]

[0077] Among them, the data point x to be predicted in each density layer data i The corresponding probability of the original category is expressed as:

[0078]

[0079] Among them, x i For a single-category dataset U m The data points to be predicted in For x i Belongs to a new category The probability of For new categories The prior probability of For a given new category Under the condition of i The probability, P(x i ) is x i The marginal probability, P(C m x i ) is x i Belongs to the original category C m The probability of t' m For a single-category dataset U m The total number of clusters generated after DBSCAN clustering;

[0080] S6. Category prediction: For the data point x to be predicted i Performs class probability calculation, selects the class with the largest probability as its predicted class label, and returns the predicted class label.

[0081] To verify the effectiveness of the above classification method, this embodiment uses precipitation data with multi-density characteristics to verify the method; in the meteorological observation at the meteorological station, the observation of meteorological elements such as temperature, air pressure, relative humidity, wind speed, wind direction, and rainfall is usually included. Among them, during 5-8 o'clock, the rain gauges at many stations will record 0.1mm precipitation. This precipitation may be real precipitation, or it may be caused by factors such as dew. The rain gauge produces precipitation records, which is usually referred to as false precipitation. At the national meteorological station, the monitoring data of the weather phenomenon instrument can be used to assist in the determination of true and false precipitation, but the general regional automatic weather station only collects several elements such as temperature, air pressure, relative humidity, wind speed, wind direction, and rainfall, and it is impossible to determine true and false precipitation. In order to distinguish the 0.1mm precipitation problem of the automatic weather station, that is, the true and false precipitation problem, we summarize the determination method of true and false precipitation in the 0.1mm precipitation data of the national meteorological station, and then promote it to the regional meteorological station. In most cases, 0.1 mm precipitation may occur in many situations. Some situations occur less frequently and have low data density, while some situations occur more frequently and have high data density. This makes true and false precipitation data have multi-density characteristics, which brings great challenges to data classification using Bayesian classifiers.

[0082] Using the classification method of this embodiment, assume that the input data set X contains five attributes, namely temperature, air pressure, relative humidity, wind speed, and wind direction. Perform Z-score standardization on each attribute in the data set to obtain the data set U = {x1, x2, ...x n}, the corresponding label is C = {c0,c1,...c n}, n means there are n data in total, each data x i ={x i,1 ,x i,2 ,...x i,5} respectively represent the observed values ​​of the five meteorological elements, and the label y i The value is 0 or 1, indicating true precipitation and false precipitation respectively.

[0083] Data segmentation: Use the original label to divide the data U into U0 and U1, which represent the real precipitation data set and the false precipitation data set respectively;

[0084] Density stratification: Use the density stratification algorithm to perform label prediction and clustering on the obtained single-category data sets U0 and U1, and obtain the corresponding density stratified data and the corresponding density stratified labels;

[0085] Through the density formula Calculate the prediction value for each data point x in the single-category dataset U0 and U1 i The data density is obtained by D0 = {D 0，1 ,D 0，2 ,...,D 0,n0} and D1={D 1，1 ,D 1，2 ,...,D 1,n1}, where n0 is the number of data points in the density data D0, n1 is the number of data points in the density data D1, K i is the data point x to be predicted in the single-category data set i The set of k data points closest to each other, E(x i ,x j ) represents the data point x i With data point x j The Euclidean distance of the data point x is 10. i The reciprocal of the average distance between the sample and the k=10 closest data points;

[0086] The density data D0 and D1 are fitted with mixed Gaussian distribution EM with model number k to calculate ICL. The ICL calculation adopts the above formula (1). The model number k is iteratively performed between 2 and T. The k0 and k1 with the maximized ICL are selected as the mixed Gaussian model number, where T is the maximum model number, which is 5 in this application.

[0087] After selecting the optimal model numbers k0 and k1, the density data D0 and D1 are predicted using the mixed Gaussian model. The mixed Gaussian model uses the above formula (2) to obtain a new density stratification label and According to the model fitting results, the data point x in the data set U is i Divide into different density layers to obtain density layer data Layers1 and Layers2;

[0088] DBSCAN clustering: Use the density clustering algorithm DBSCAN to cluster the density-layered data. The density clustering algorithm DBSCAN uses two parameters, namely the neighborhood radius ε and the minimum number of neighbors MinPts. In the application, ε is taken as 3 and MinPts is taken as 6 to generate a new category label C0={c 0，1 ,c 0，2 ,...c 0,m1} and C1={c 1，1 ,c 1，2 ,…c 1,m2}, where m1 is the total number of clusters generated after DBSCAN clustering of the single-category dataset U0, and m2 is the total number of clusters generated after DBSCAN clustering of the single-category dataset U1;

[0089] Bayesian model fitting: Based on the new category labels, the Bayesian model is fitted using the Bayesian formula. The above Bayesian formulas (3) and (4) are used in this application.

[0090] Category probability calculation: The probability of the new category corresponding to the data point x in each density layered data in the single category data U0 and U1 is calculated by the above formula (6). and Then, the probability of x being true precipitation or false precipitation is calculated using formula (7).

[0091] Example 2

[0092] like Figure 2 and 3 As shown, this embodiment provides a multi-density data classification system of density hierarchical clustering, including:

[0093] Data segmentation module: obtains labeled data sets, divides the original data using the original labels, and divides the data with the same labels into the same data set, thereby obtaining multiple single-category data sets and corresponding category labels;

[0094] Density stratification module: Use density stratification algorithm to perform label prediction and clustering on the obtained single-category data sets to obtain corresponding density stratification data and corresponding density stratification labels;

[0095] The density stratification module includes:

[0096] Density data calculation module: calculates the density data of each single-category data set through the density formula;

[0097] Model fitting and selection module: Model fitting and optimal model number selection based on calculated density data;

[0098] Density layer prediction module: After selecting the optimal number of models, the Gaussian mixture model is used to predict the density data to obtain density layer labels; the original data points are divided into different density layers according to the model fitting results to obtain density layer data;

[0099] DBSCAN clustering module: Use the density clustering algorithm DBSCAN to cluster each data layer in the obtained density stratified data and generate new category labels;

[0100] Bayesian model fitting module: Use new category labels and Bayesian formula to fit the Bayesian model for each single-category data set to achieve data classification; Use the Bayesian formula to fit the Bayesian model to use different Bayesian classifiers based on different prior assumptions, such as Naive Bayes, Gaussian Naive Bayes, etc.;

[0101] Category probability calculation module: Use the fitted Bayesian model to calculate the data points to be predicted in each single-category data set, calculate the probability that the data points to be predicted belong to the new category, and calculate the probability that the data points to be predicted belong to the original category;

[0102] Category prediction module: performs category probability calculation for the data point to be predicted, selects the category with the largest probability as its predicted classification label, and returns the predicted category label.

[0103] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, devices or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0104] The present invention is described with reference to flowcharts and / or block diagrams of methods and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0105] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0106] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0107] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A multi-density data classification method based on density hierarchical clustering, characterized in that: The following steps are involved: Obtain a data set with category labels, divide the original data using the original category labels, and divide the data with the same category labels into the same data set, thereby obtaining multiple single-category data sets and corresponding category labels; The obtained single-category data sets are respectively subjected to label prediction and clustering using density stratification algorithms to obtain corresponding density stratified data and corresponding density stratified labels; Each data layer in the obtained density stratified data is clustered using the density clustering algorithm DBSCAN to generate new category labels; Use the new category label and Bayesian formula to fit the Bayesian model for each single-category data set to achieve data classification; The fitted Bayesian model is used to calculate the probability that the data point to be predicted belongs to the new category in each single-category data set, and then the probability that the data point to be predicted belongs to the original category is calculated; Perform category probability calculation for the data point to be predicted, select the category with the largest probability as its predicted classification label, and return the predicted category label.

2. The multi-density data classification method based on density hierarchical clustering according to claim 1 is characterized in that: Using the density stratification algorithm on a single-class dataset involves the following steps: Through the density formula Calculate the single-category dataset U m For each data point x to be predicted i The data density is obtained by m , where K i is the data point x to be predicted in the single-category data set i The set of k data points closest to each other, E(x i ,x j ) represents the data point x i With data point x j The Euclidean distance of For density data D m Perform EM fitting of a mixed Gaussian distribution with model number k to calculate ICL. This operation is performed iteratively between 2 and T, and the k with the maximum ICL is selected. m As the number of mixed Gaussian models, where T is the maximum number of models; Select the optimal number of models k m Then, the density data D is analyzed using the mixed Gaussian model. m Make predictions and get new density stratified labels According to the model fitting results, the original data is divided into different density layers to obtain density layered data Layers m .

3. The multi-density data classification method based on density hierarchical clustering according to claim 2 is characterized in that: The data density is for each data point x i The inverse of the average distance to the k closest data points.

4. The multi-density data classification method based on density hierarchical clustering according to claim 2 is characterized in that: The output of EM in the mixed Gaussian distribution EM fitting is a specified number of Gaussian model parameters {Θ1, Θ2, ...}, and the calculation formula of ICL is as follows: in, is the degree of fit of the model, V m,K is the number of free parameters in the mixture model m containing K distributions, n m For a single-category dataset U m The number of data points to be predicted in .

5. The multi-density data classification method based on density hierarchical clustering according to claim 2 is characterized in that: The Gaussian mixture model is expressed as: Among them, H is the number of data density layers, Θ h is the parameter of the h-th Gaussian model, α h is the weight of the i-th Gaussian model, N(ρθ h ) is the probability density function of the h-th Gaussian model, 6. The multi-density data classification method based on density hierarchical clustering according to claim 1 is characterized in that: The density clustering algorithm DBSCAN uses two parameters, namely the neighborhood radius ε and the minimum number of neighbors MinPts.

7. The multi-density data classification method based on density hierarchical clustering according to claim 1 is characterized in that: The Bayesian formula used to fit the Bayesian model is: Among them, P(C m a) is to calculate the new input feature vector α in category C m The concept of the posterior under P(C m ) is a priori concept, P(C m A i ) is feature A i In Category C m The kernel probability density function under i,j C m ) is the conditional probability density function of Gaussian naive Bayes, μ j and Category C m The mean and variance of the jth feature, α i,j is the value of the jth feature of the i-th sample; P(a i |C m ) is the decision function of the multivariate Gaussian Bayes, μ m and Σ m C m The mean vector and covariance matrix of the class attributes, α i is the feature vector of the i-th sample, and p is the number of features.

8. The multi-density data classification method based on density hierarchical clustering according to claim 1 is characterized in that: The data points x to be predicted in each density layer data i The corresponding probability of the new category is expressed as: The data points x to be predicted in each density layer data i The corresponding probability of the original category is expressed as: Among them, x i For a single-category dataset U m The data points to be predicted in For x i Belongs to a new category The probability of For new categories The prior probability of For a given new category Under the condition of i The probability, P(x i ) is x i The marginal probability, P(C m x i ) is x i Belongs to the original category C m The probability of t' m For a single-category dataset U m The total number of clusters generated after DBSCAN clustering.

9. A multi-density data classification system based on density hierarchical clustering, characterized in that: include: Data segmentation module: obtains labeled data sets, divides the original data using the original labels, and divides the data with the same labels into the same data set, thereby obtaining multiple single-category data sets and corresponding category labels; Density stratification module: Use density stratification algorithm to perform label prediction and clustering on the obtained single-category data sets to obtain corresponding density stratification data and corresponding density stratification labels; DBSCAN clustering module: Use the density clustering algorithm DBSCAN to cluster each data layer in the obtained density stratified data and generate new category labels; Bayesian model fitting module: Use new category labels and Bayesian formulas to fit the Bayesian model for each single-category data set to achieve data classification; Category probability calculation module: Use the fitted Bayesian model to calculate the data points to be predicted in each single-category data set, calculate the probability that the data points to be predicted belong to the new category, and calculate the probability that the data points to be predicted belong to the original category; Category prediction module: performs category probability calculation for the data point to be predicted, selects the category with the largest probability as its predicted classification label, and returns the predicted category label.

10. The multi-density data classification system based on density hierarchical clustering according to claim 9, characterized in that: The density stratification module comprises: Density data calculation module: calculates the density data of each single-category data set through the density formula; Model fitting and selection module: Model fitting and optimal model number selection based on calculated density data; Density layer prediction module: After selecting the optimal number of models, the Gaussian mixture model is used to predict the density data to obtain density layer labels; the original data points are divided into different density layers according to the model fitting results to obtain density layer data.

Images