Uncertain data classification method for Internet of Things based on trust function and evidence correction

Abstract

This invention provides a method for classifying uncertain data in the Internet of Things (IoT) based on a trust function and evidence correction. The method includes: modeling the uncertain data of a sample dataset according to a set uncertainty, obtaining an uncertain dataset; dividing the uncertain dataset into a test set and a training set; obtaining various data feature dimensions from the sample dataset, constructing multiple classifiers by pairwise combinations of these data feature dimensions, and calculating the Mahalanobis distance between the sample to be classified and the training samples under each classifier; applying the K-NN classification algorithm to weight the Mahalanobis distance to generate evidence, and correcting all evidence using a cosine similarity matrix to obtain corrected evidence; making a decision and identification based on the corrected evidence to obtain the classification result of the sample to be classified. This invention uses Mahalanobis distance as a distance metric and applies the DS combination rule to fuse multiple pieces of evidence to make a classification decision. It solves the problem of low accuracy in classifying uncertain data.

Description

Technical Field

[0001] This invention relates to the field of uncertain data classification, and more particularly to an IoT uncertain data classification method based on trust functions and evidence correction. Background Technology

[0002] With the rise of the Internet of Things (IoT), data collection is constantly evolving and becoming more innovative. Analyzing sensor data can enable the "Internet of Everything," showing promise in fields such as industry, agriculture, and healthcare. However, in real-world engineering, uncertainties exist in the data due to factors such as noise, transmission, and sensor characteristics. The formation of uncertain data is unavoidable, and in-depth exploration of the field of uncertain data is a current research hotspot.

[0003] Since probability theory is difficult to reliably model and analyze uncertain data, evidence theory, as an extension of it, can construct a theoretical framework for reasoning about uncertain information. To address the problem of reasoning about uncertainty, multi-source data information can be used as evidence, and more credible classification decision results can be obtained after evidence fusion. Summary of the Invention

[0004] Embodiments of the present invention provide a method for classifying uncertain IoT data based on a trust function and evidence correction, so as to effectively classify uncertain IoT data.

[0005] To achieve the above objectives, the present invention adopts the following technical solution.

[0006] IoT uncertain data classification methods based on trust functions and evidence correction include:

[0007] Data samples in the Internet of Things (IoT) scenario are acquired, a sample dataset is established, and uncertain data modeling is performed on the sample dataset according to the set uncertainty to obtain an uncertain dataset. A portion of the uncertain dataset is used as the test set of the sample dataset, and the other data in the uncertain dataset is used as the training set.

[0008] Obtain each data feature dimension from the sample dataset, construct multiple classifiers by pairwise combinations of the data feature dimensions, select samples to be classified in the test set, and calculate the Mahalanobis distance between the samples to be classified and the training samples in the training set for each classifier:

[0009] Under each classifier, the K-NN classification algorithm is applied to generate evidence by weighting the Mahalanobis distances corresponding to the samples to be classified. The cosine similarity matrix is ​​then used to correct all the evidence corresponding to the samples to be classified, resulting in the corrected evidence.

[0010] Decisions and identifications are made based on the revised evidence to obtain the classification results of the samples to be classified.

[0011] Preferably, the process of acquiring data samples in an IoT scenario, establishing a sample dataset, and performing uncertain data modeling on the sample dataset based on a set uncertainty to obtain an uncertain dataset includes:

[0012] Acquire data samples from IoT scenarios, establish a sample dataset, model the uncertainty of the sample dataset, and identify the characteristics ω of uncertain data within the entire sample dataset. j Minimum value is The maximum value is Then the feature ω of the i-th sample i,j The uncertainty interval is:

[0013]

[0014] Where U represents the uncertainty of the uncertain dataset, and rand1 and rand2 are two random numbers generated on [0,1].

[0015] Assume the attribute ω of the i-th object i,j Uncertain interval V i,j =[a i,j ,b i,j If ω is the attribute of the i-th object, then... i,j sampling point v i,j for:

[0016] v i,j =a i,j +(b i,j -a i,j )×rand3

[0017] Where: rand3 is a random number generated in [0,1];

[0018] Generate an uncertain interval V with uncertainty U i,j =[a i,j ,b i,j ]for:

[0019] V i,j =[20-(20-1)×U×rand1,20+(50-20)×U×rand2]

[0020] The uncertain value is: v i,j =a i,j +(b i,j -a i,j )×rand3

[0021] The above processing steps are used to model uncertain data in the sample dataset.

[0022] Preferably, the step of selecting samples to be classified in the test set and calculating the Mahalanobis distance between the samples to be classified and the training samples in the training set for each classifier includes:

[0023] Select samples to be classified in the test set, and calculate the Mahalanobis distance between the samples to be classified and the training samples in the training set for each classifier:

[0024]

[0025] Where Σ is the covariance matrix, which is generated only when the number of samples is greater than the number of dimensions. Mahalanobis distance is used, the sample to be classified is x, and the training sample is y.

[0026] Preferably, the step of generating evidence by weighting the Mahalanobis distances corresponding to the samples to be classified using the K-NN classification algorithm under each classifier includes:

[0027] Step 3.1: For each classifier, select the K sample points with the smallest Mahalanobis distance, denoted as {d}. ω,1 ,d ω,2 ,…,d ω,k};

[0028] Step 3.2: Obtain the category labels of the K nearest neighbors {l ω,1 ,l ω,2 ,…,l ω,k};

[0029] Step 3.3: Use the distance weighting function y = tanh(-x) + 1 to weight the distances of the K sample points with the smallest Mahalanobis distance;

[0030] Step 3.4: Normalize the weighted result and denote it as the probability {p}. ω,1 ,p ω,2 ,…,p ω,k Based on the K-nearest neighbor category labels, the probabilities of the same category are accumulated to obtain one piece of evidence corresponding to the sample to be classified output by the classifier;

[0031] The above processing procedure is performed for each classifier to obtain one piece of evidence corresponding to the sample to be classified output by each classifier.

[0032] Preferably, the step of using a cosine similarity matrix to correct all evidence corresponding to the sample to be classified, to obtain corrected evidence, includes:

[0033] Step 4.1: Calculate the cosine similarity matrix S among all the evidence corresponding to the sample to be classified:

[0034]

[0035] The cosine similarity between evidence m1 and evidence m2 is:

[0036]

[0037] Step 4.2: Summing the cosine similarity matrix S along the same lines yields the trust support matrix B.

[0038]

[0039] Where N is the number of classifiers, i.e., the total number of pieces of evidence;

[0040] Step 4.3: Use matrix B to weight all the evidence corresponding to the sample to be classified, and then normalize it to obtain the corrected evidence.

[0041] Preferably, the step of making decisions and identifications based on the corrected evidence to obtain the classification results of the samples to be classified includes:

[0042] Step 5.1: Use the DS combination rule to fuse all the corrected evidence corresponding to the sample to be classified:

[0043]

[0044]

[0045] κ is a quantitative parameter for the degree of conflict between pieces of evidence. The DS combination rule is only effective when κ < 1, i.e. when the evidence is not completely conflicting.

[0046] Step 5.2: Take the category w with the highest pignistic probability in the fusion result as the final classification result of the sample to be classified.

[0047] As can be seen from the technical solutions provided by the embodiments of the present invention above, the IoT uncertain data classification method based on trust function and evidence correction provided by the present invention uses Mahalanobis distance as the distance metric, selects a distance weighting function to construct evidence, uses cosine similarity to correct the evidence according to the degree of conflict, and finally applies the DS combination rule to complete the fusion of multiple pieces of evidence and make a classification decision. This solves the problem of uncertain data classification accuracy.

[0048] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and will become apparent from the description or may be learned by practice of the invention. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart illustrating a method for classifying uncertain IoT data based on trust functions and evidence correction, provided in an embodiment of the present invention. Detailed Implementation

[0051] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0052] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or couplings. The term “and / or” as used herein includes any and all combinations of one or more of the associated listed items.

[0053] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0054] To facilitate understanding of the embodiments of the present invention, the following will provide further explanation and description with reference to the accompanying drawings and several specific embodiments. These embodiments do not constitute a limitation on the embodiments of the present invention.

[0055] This invention provides a processing flow for an IoT uncertain data classification method based on trust functions and evidence correction, as follows: Figure 1As shown, the processing steps include the following:

[0056] Step 1: Obtain data samples from IoT scenarios and establish a sample dataset:

[0057] In this embodiment of the invention, uncertainty modeling is performed on the input dataset Blood Transfusion to simulate an Internet of Things dataset.

[0058] The Blood Transfusion dataset has a sample size of 748, divided into a blood donor category (570 records) and a non-donor category (178 records). The class label indicates whether the person donated blood in March 2007. "1" indicates that the person donated blood, and "0" indicates that the person did not donate blood.

[0059] In the entire Blood Transfusion dataset, the feature ω of uncertain data j Minimum value is The maximum value is Then the feature ω of the i-th sample i,j The uncertainty interval is:

[0060]

[0061] Where U represents the uncertainty of the uncertain dataset, and rand1 and rand2 are two random numbers generated on [0,1].

[0062] Assume the attribute ω of the i-th object i,j Uncertain interval V i,j =[a i,j ,b i,j If ω is the attribute of the i-th object, then... i,j sampling point v i,j for:

[0063] v i,j =a i,j +(b i,j -a i,j )×rand3

[0064] Where: rand3 is a random number generated in [0,1].

[0065] In the Blood Transfusion dataset selected in this embodiment of the invention, the attribute "total number of blood donations (times)" is chosen. The maximum value of this attribute is 50, and the minimum value is 1. The fourth data object has a "total number of blood donations (times)" value of 20. Therefore, an uncertainty interval V with uncertainty U is generated. i,j =[a i,j ,b i,j ]for:

[0066] V i,j =[20-(20-1)×U×rand1,20+(50-20)×U×rand2]

[0067] The uncertain value is: v i,j =a i,j +(b i,j -a i,j )×rand3.

[0068] The above method can be used to model uncertain data in a dataset. The uncertainty U can typically be selected as 0.01, 0.05, 0.10, or 0.20. Table 1 shows a comparison of the Blood Transfusion dataset before and after modeling when the uncertainty U is 0.01:

[0069] Table 1 Comparison of Blood Transfusion dataset before and after modeling

[0070]

[0071]

[0072] Step 2: Obtain feature dimensions from the IoT dataset and divide it into test and training sets:

[0073] Obtain the feature dimensions from the IoT dataset; the Blood Transfusion dataset has 4 feature dimensions. Construct multiple (e.g., 6) classifiers by pairwise combinations of these feature dimensions. The output of each classifier serves as evidence for the sample's class. Use a portion of the uncertain dataset as the test set and the remaining data as the training set; for example, 25% of the dataset as the test set and 75% as the training set. Select samples to be classified from the test set and calculate the Mahalanobis distance between the samples to be classified and the training samples for each classifier.

[0074]

[0075] Where Σ is the covariance matrix. Mahalanobis distance can only be used to generate a covariance matrix when the number of samples is greater than the number of dimensions. The sample to be classified is x, and the training sample is y.

[0076] Step 3: Apply the K-NN classification algorithm to generate evidence after distance weighting.

[0077] Step 3.1: For each classifier, select the K sample points with the smallest Mahalanobis distance, denoted as {d}. ω,1 ,d ω,2 ,…,dω,k Taking K=13, the 13 minimum Mahalanobis distances selected by classifier 1 are shown in the table below:

[0078] Table 2 shows the K minimum Mahalanobis distances for classifier 1.

[0079]

[0080] Step 3.2: Obtain the category labels of the K nearest neighbors {l ω,1 ,l ω,2 ,…,l ω,k By obtaining the sample index, the category label can be found.

[0081] Table 2 shows the indices of the K sample points for classifier 1.

[0082]

[0083]

[0084] Step 3.3: Use the distance weighting function y = tanh(-x) + 1 to weight the distances of the K sample points with the smallest Mahalanobis distance.

[0085] Step 3.4: The normalized value described in Step 3.3 is denoted as probability {p}. ω,1 ,p ω,2 ,…,p ω,k Based on the K nearest neighbor class labels, the probabilities of the same class are summed to obtain the output of classifier ω:

[0086] (0.4896032527458591, 0.510396747254141)

[0087] As evidence m1.

[0088] We obtain one piece of evidence for each classifier for the sample to be classified.

[0089] Step 4: Correct the evidence using the cosine similarity matrix:

[0090] Includes the following sub-steps:

[0091] Step 4.1: Calculate the cosine similarity matrix S among all pieces of evidence:

[0092]

[0093] The cosine similarity between m1 and m2 is:

[0094]

[0095] m1 and m2 represent two pieces of evidence.

[0096] Step 4.2: Summing the cosine similarity matrix S along the same lines yields the trust support matrix B.

[0097]

[0098] In the formula, N represents the number of classifiers, i.e., the total number of pieces of evidence.

[0099] Step 4.3: Weight all the evidence using matrix B and normalize it to obtain the corrected evidence. Each corrected piece of evidence can be seen as a preliminary judgment result of each classifier regarding the sample category. Therefore, it is necessary to fuse and make judgment decisions on the multiple corrected pieces of evidence to obtain a more comprehensive and reliable classification result for the sample to be classified.

[0100] Step 5: Decision-making and identification, determining the classification result.

[0101] Includes the following sub-steps:

[0102] Step 5.1: Fuse the evidence revised in Step 4 using the DS combination rule:

[0103]

[0104]

[0105] κ is a quantitative parameter for the degree of conflict between pieces of evidence. The DS combination rule is only effective when κ < 1, i.e. when the evidence is not completely conflicting.

[0106] Step 5.2: The category corresponding to the highest probability in the fusion result is the classification result. In the evidence theory framework, the pignistic probability is the most commonly used probability transformation method. This invention also adopts the pignistic probability formula to convert the basic trust function of evidence into the probability metric BetP. m :

[0107]

[0108] Where |A| represents the number of elements contained in set A.

[0109] Therefore, the category w with the highest pignistic probability in the fusion result is the final classification result. Furthermore, by outputting the classification results of the samples to be classified, the classification accuracy of this method can be obtained.

[0110] The table below compares our method with other uncertain classification algorithms. As can be seen from the table, our method has a higher accuracy than other methods.

[0111] Table 3 Comparison of accuracy of four classification algorithms under different levels of uncertainty.

[0112]

[0113] In summary, the present invention provides an IoT uncertain data classification method based on a trust function and evidence correction. By using Mahalanobis distance as the distance metric, it calculates the distance by combining data dimensions pairwise. This makes the distance independent of dimensions, eliminating the need to know whether the chosen data unit is appropriate. It also takes into account the data distribution and potential correlations between dimensions.

[0114] This invention provides an IoT uncertain data classification method based on trust function and evidence correction. It uses the hyperbolic tangent function to map Mahalanobis distance to the trust function, and based on the weighted approach, assigns weights according to the importance of each piece of evidence, making the final fusion result more accurate.

[0115] The present invention provides an IoT uncertain data classification method based on trust function and evidence correction. It uses cosine similarity to correct evidence and applies DS combination rules to perform self-fusion on the basis of new evidence. The final fusion result is more accurate and effectively solves the impact of conflicting evidence on the fusion result.

[0116] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of one embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing the present invention.

[0117] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus necessary general-purpose hardware platforms. 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 can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present invention.

[0118] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for apparatus or system embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The apparatus and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. 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.

[0119] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A classification method for uncertain IoT data based on trust function and evidence correction, characterized in that, include: Data samples in the Internet of Things (IoT) scenario are acquired, a sample dataset is established, and uncertain data modeling is performed on the sample dataset according to the set uncertainty to obtain an uncertain dataset. A portion of the uncertain dataset is used as the test set of the sample dataset, and the other data in the uncertain dataset is used as the training set. Obtain each data feature dimension from the sample dataset, construct multiple classifiers by pairwise combinations of the data feature dimensions, select samples to be classified in the test set, and calculate the Mahalanobis distance between the samples to be classified and the training samples in the training set for each classifier: Under each classifier, the K-NN classification algorithm is applied to generate evidence by weighting the Mahalanobis distances corresponding to the samples to be classified. The cosine similarity matrix is ​​then used to correct all the evidence corresponding to the samples to be classified, resulting in the corrected evidence. Make decisions and identify based on the revised evidence, and obtain the classification results of the sample to be classified; The method of generating evidence by weighting the Mahalanobis distances corresponding to the samples to be classified using the K-NN classification algorithm under each classifier includes: Step 3.1: For each classifier, select the K sample points with the smallest Mahalanobis distance, denoted as {d}. ω,1 ,d ω,2 ,…,d ω,k }; Step 3.2: Obtain the category labels of the K nearest neighbors {l ω,1 ,l ω,2 ,…,l ω,k }; Step 3.3: Use the distance weighting function y = tanh(-x) + 1 to weight the distances of the K sample points with the smallest Mahalanobis distance; Step 3.4: Normalize the weighted result and denote it as the probability {p}. ω,1 ,p ω,2 ,…,p ω,k Based on the K-nearest neighbor category labels, the probabilities of the same category are accumulated to obtain one piece of evidence corresponding to the sample to be classified output by the classifier; The above processing procedure is performed for each classifier to obtain one piece of evidence corresponding to the sample to be classified output by each classifier. The method of using a cosine similarity matrix to correct all evidence corresponding to the sample to be classified, resulting in corrected evidence, includes: Step 4.1: Calculate the cosine similarity matrix S among all the evidence corresponding to the sample to be classified: The cosine similarity between evidence m1 and evidence m2 is: Step 4.2: Summing the cosine similarity matrix S along the same lines yields the trust support matrix B. Where N is the number of classifiers, i.e., the total number of pieces of evidence; Step 4.3: Use matrix B to weight all the evidence corresponding to the sample to be classified, and then normalize it to obtain the corrected evidence.

2. The method according to claim 1, characterized in that, The process of acquiring data samples in an IoT scenario, establishing a sample dataset, and performing uncertain data modeling on the sample dataset based on a set uncertainty to obtain an uncertain dataset includes: Acquire data samples from IoT scenarios, establish a sample dataset, model the uncertainty of the sample dataset, and identify the characteristics ω of uncertain data within the entire sample dataset. j Minimum value is The maximum value is Then the feature ω of the i-th sample i,j The uncertainty interval is: Where U represents the uncertainty of the uncertain dataset, and rand1 and rand2 are two random numbers generated on [0,1]. Assume the feature ω of the i-th sample i,j Uncertain interval V i,j =[a i,j ,b i,j If ω is the attribute of the i-th object, then... i,j sampling point v i,j for: v i,j =a i,j +(b i,j -a i,j )×rand3 Where: rand3 is a random number generated in [0,1]; Generate an uncertain interval V with uncertainty U i,j =[a i,j ,b i,j ]for: In i,j =[20-(20-1)×U×rand1,20+(50-20)×U×rand2] The uncertain value is: v i,j =a i,j +(b i,j -a i,j )×rand3 The above processing steps are used to model uncertain data in the sample dataset.

3. The method according to claim 2, characterized in that, The process of selecting samples to be classified in the test set and calculating the Mahalanobis distance between the samples to be classified and the training samples in the training set for each classifier includes: Select samples to be classified in the test set, and calculate the Mahalanobis distance between the samples to be classified and the training samples in the training set for each classifier: Where Σ is the covariance matrix, which is generated only when the number of samples is greater than the number of dimensions. Mahalanobis distance is used, the sample to be classified is x, and the training sample is y.

4. The method according to claim 1, characterized in that, The aforementioned decision-making and identification based on the revised evidence, and obtaining the classification result of the sample to be classified, includes: Step 5.1: Use the DS combination rule to fuse all the corrected evidence corresponding to the sample to be classified: κ is a quantitative parameter for the degree of conflict between pieces of evidence. The DS combination rule is only effective when κ < 1, i.e. when the evidence is not completely conflicting. Step 5.2: Take the category w with the highest pignistic probability in the fusion result as the final classification result of the sample to be classified.

Images