Data mining method based on fuzzy monotone sum entropy

Abstract

The invention discloses a data mining method based on fuzzy monotone and entropy, and relates to the technical field of data processing. The invention provides a data mining method based on fuzzy monotone and entropy, which utilizes entropy to remap a relationship between continuous features and discrete decision classes, and measures feature importance by analyzing fuzzy monotone membership change of the entropy. In addition, a variable coefficient is introduced, the change condition of the fuzzy monotonic membership degree is intuitively and effectively measured, and parameterization setting is carried out on the change condition of the fuzzy monotonic membership degree value. Compared with the prior art, the problem that an existing fuzzy monotone relation method based on inclusion degree cannot directly analyze a classification task data set is solved, and the problems that a distance method adopted by feature selection based on a neighborhood rough set framework is low in calculation efficiency and classification accuracy and prone to being interfered by data noise are solved.

Description

Technical Field

[0001] The invention relates to the technical field of data processing, and in particular to a data mining method based on fuzzy monotony and entropy. Background Art

[0002] With the continuous development of data collection and storage technology, massive amounts of data are collected and stored as large data sets by researchers. However, redundant features and incomplete features are common problems of large data sets. Such features as input will increase the computational complexity of the classifier, resulting in increased classifier execution time and decreased classification accuracy. In order to solve such problems, feature selection technology has been proposed and widely used in recent years. Feature selection is a technology that reduces redundant input features while ensuring the classification accuracy of the classifier. The purpose is to retain the feature attributes that are most relevant to the decision attributes and reduce the input of redundant attributes, thereby achieving the dual purpose of reducing the execution time of the classifier and improving the generalization of the classifier.

[0003] In order to process continuous features, the neighborhood rough set framework was proposed as an extension of the rough set. Hu et al. combined the neighborhood concept with mutual information [Measuring relevance between discrete and continuous features based on neighborhood mutual information], proposed the measure of neighborhood mutual information, and solved the problem that information entropy needs to be discretized. The introduction of the neighborhood concept helps rough sets and mutual information to directly process numerical features and avoid discretization. Inspired by Hu, Wang et al. introduced the neighborhood into the decision theory rough set model (DTRS) to directly process numerical features [Neighborhood based decision theoretic rough set models]. In order to speed up the reduction of neighborhood rough sets, Liu et al. designed a fast reduction algorithm based on neighborhood rough sets [Quick attribute reduction algorithm for neighborhood rough set model], which accelerated the calculation of the size of the positive region by using the hash-based storage bucket partitioning method. Facing the problem of unbalanced data distribution, Wang et al. combined the advantages of Hu's knn neighborhood rough set [Mixed feature selection based on granulation and ap proximity] and δ neighborhood rough set [Neighborhood rough set based heterogeneous feature subset selection] algorithms and proposed k-nearest neighbor neighborhood rough set [Attribute reduction based on k-nearest neighborhood rough sets] to solve the problem of sparse and dense data feature distribution to a certain extent. Later, Hu. Meng et al. defined the concept of overlap based on Wang's work, which helped the k-nearest neighbor neighborhood rough set filter redundant attributes and accelerated the convergence speed of the k-nearest neighbor neighborhood rough set. In order to further improve the computational efficiency of neighborhood rough sets, Hu et al. used attribute importance as the screening criterion for heuristic search and proposed a feature selection algorithm based on fast search and neighborhood rough sets [Efficient Symbolic and Numerical Attribute Reduction with Neighborhood Rough Sets].Using the positive correlation between the size of positive regions and the number of attributes, the repeated calculation of positive regions is reduced. Based on Hu et al., Xia introduced granular computing and proposed the GBNRS model [GBNRS: A Novel Rough Set Algorithm for Fast Adaptive Attribute Reduction in Classification]. This model introduced granular computing, adaptively set the neighborhood size for each object, and continuously fit the upper and lower approximations through the granularity ball. Compared with other neighborhood algorithms, it has a high degree of adaptability. It realizes the neighborhood rough set without parameter setting. These algorithms all use distance as a metric to measure the similarity or distance between samples. Different from these algorithms, Liang proposed a correlation analysis method based on fuzzy monotone theory [A Novel Fuzzy Monotone Rela tionship Method Withits Application on Inclusion Degree] to analyze the correlation between continuous features, sort the continuous features and decision classes, and analyze the fuzzy monotone degree to analyze the correlation between features and decision classes.

[0004] In addition, the existing Chinese patent CN 102609469A discloses a fuzzy rough monotone data mining method based on inclusion, which rearranges the decision attributes and conditional attributes by value, then divides the rearranged set into intervals, and then sets judgment rules according to the membership function of each interval and the inclusion degree of each interval to judge the relationship between the decision attributes and the conditional attributes, thereby establishing a fuzzy inclusion monotone dependency relationship model; the preliminary relationship between the decision attributes and the conditional attributes is mined through the relationship model, and the decision filtering rules are set to obtain the data set and optimal data for the simplified conditional attributes.

[0005] Then the above technology still has the following problems:

[0006] 1. The fuzzy monotonic relationship method based on inclusion degree cannot directly perform attribute simplification or correlation analysis on the data set whose classification target, i.e., the decision class, is discrete data;

[0007] 2. The neighborhood rough feature selection method based on distance is easily disturbed when the sample data is unevenly distributed, and the high-dimensional data processing efficiency is not high, and the classification accuracy is not high. The reasons are:

[0008] (1) The fuzzy monotonic relationship method based on inclusion requires sorting the two features separately to obtain a new sample sorting. However, the sample sequence does not change much before and after the discrete decision class sorting, so it is impossible to directly perform fuzzy inclusion monotonic analysis.

[0009] (2) Feature selection based on the neighborhood rough set framework uses distance as a metric, which requires frequent calculation of distances between samples. The amount of calculation is large, and when the sample distribution is uneven, it is easily disturbed, resulting in low accuracy in high-dimensional data classification.

[0010] In order to solve the above problems, the present invention proposes a data mining method based on fuzzy monotony and entropy. Summary of the invention

[0011] The purpose of the present invention is to propose a data mining method based on fuzzy monotonicity and entropy to solve the problem that the existing fuzzy monotonic relationship method based on inclusion cannot directly analyze the classification task data set, and the distance method used in feature selection based on the neighborhood rough set framework has low computational efficiency and classification accuracy and is easily disturbed by data noise.

[0012] In order to achieve the above object, the present invention adopts the following technical solutions:

[0013] A data mining method based on fuzzy monotone and entropy, including the following contents:

[0014] S1. Normalize and preprocess the original features in the data set, construct a training data set, and initialize the candidate feature reduction subset to an empty set;

[0015] S2, traverse the attribute features in the training data set, sort the continuous variable attribute features in ascending order to obtain a sample sequence of the decision class;

[0016] S3. According to the principle of interval division, the sample sequence is partitioned, the average value of the entropy values ​​of different intervals is calculated, and the interval entropy value table of the feature is obtained;

[0017] S4, sorting the interval entropy value table obtained in S3 in ascending order, and calculating the fuzzy monotonic membership value of the entropy value table;

[0018] S5, calculating the coefficient of variation of the fuzzy monotone membership value, if it is less than the set threshold, adding the corresponding feature to the feature reduction subset;

[0019] S6, check whether the features have been traversed. If not, return to S2 to continue iterating. If all features have been traversed, sort them according to the maximum value of the fuzzy monotone membership value in the entropy table of each feature, and output the reduced feature subset.

[0020] S7. Input the features in the feature reduction subset in S6 into the classifier one by one for iterative training, and output the optimal feature subset with the highest accuracy and the classification accuracy.

[0021] Preferably, S3 specifically includes the following contents:

[0022] The sample sequence of the decision class is recorded as C, and the sample sequence C is partitioned. Assume that the number of elements in the interval is m, and the value range of m is [2, n / 2], where n represents the number of samples; traverse the value of m and obtain:

[0023] C={C1,C2,…,C k}

[0024] Among them, k = n / m; C1 represents an interval, and the others are similar;

[0025] Calculate each interval C k The entropy value is used to describe the degree of confusion of interval decision making, and the average value of the entropy value is further calculated to obtain the average value table of the entropy value under different k values. The specific calculation formula is:

[0026] H(X)=-∑ x p(x)logp(x)

[0027]

[0028] Among them, H(X) represents the entropy value of each interval; H(D k ′) represents the average entropy value; p(x) represents the probability of sample x.

[0029] Preferably, the S4 specifically includes the following contents:

[0030] The initial sequence of the entropy average value table is recorded as A, and the sequence of the entropy average value table after ascending sorting is recorded as B. Sequence A and sequence B are distinguished. Assume that the number of elements in the interval is q, and the value range of q is [2,n′ / 2], where n′ represents the sequence length;

[0031] Under different q values, the fuzzy monotonic membership of the entropy value is calculated according to the fuzzy monotonic membership formula. The specific calculation formula is:

[0032]

[0033] Among them, A j represents the initial sequence in the entropy average table, C i It represents the sorted sequence, U Av and U Cv Represents each partition of the sequence, v represents the subscript of the interval, and the mathematical symbol It means to take the conjunction of the membership values ​​of two intervals. Here, there are two fuzzy monotony cases: fuzzy monotony decreasing and fuzzy monotony increasing. In the present invention, only fuzzy monotony increasing is cited as an example. Fuzzy monotony decreasing is sorted in descending order in the interval sorting, and vice versa.

[0034] Preferably, the S5 specifically includes the following contents:

[0035] Analyze the changes of fuzzy monotonic membership and observe the fluctuation of fuzzy monotonic membership value through the coefficient of variation. The calculation formula of the coefficient of variation is as follows:

[0036]

[0037] Among them, fm i is a single fuzzy membership value corresponding to a certain q value; μ represents the average value of the fuzzy membership value in the entropy table; It means to calculate the mean of the subset. Calculate the standard deviation of the subset;

[0038] If the coefficient of variation is less than the set threshold θ, it means that the feature has a fuzzy monotonically increasing relationship with the decision class. The maximum value of the fuzzy monotonous membership value is recorded and the feature is added to the feature reduction subset.

[0039] Preferably, the classifier in S7 specifically refers to a support vector machine classifier.

[0040] Compared with the prior art, the present invention provides a data mining method based on fuzzy monotony and entropy, which has the following beneficial effects:

[0041] (1) The present invention introduces entropy values ​​to describe the degree of confusion of decision classes after feature sorting, maps the fuzzy monotonic relationship between features and decision classes to a feature entropy value table for analysis, and outputs classified decision data for input continuous feature data. It can be applied to continuous input data sets of classification tasks, does not require discretization of input continuous data, does not use distance metrics, has good robustness, and solves the problem that fuzzy monotonicity cannot be directly applied to feature selection of output decision classification.

[0042] (2) The present invention uses the coefficient of variation to describe the fluctuation of the fuzzy monotone membership value, the feature selection method of relative distance and other metrics, which requires fewer parameters and has better interpretability. According to the fuzzy monotone references, the fuzzy monotone theory is mainly based on the energy conservation theory. The energy conservation and entropy theory are both the most basic and important theories to explain the world. The parameterization setting of the present invention is simple and more interpretable, and it can better screen features with good correlation.

[0043] (3) The present invention uses fuzzy monotonic membership value as a metric to analyze the fuzzy monotonicity of features and decision classes, and analyzes the correlation between features and decision classes through the fuzzy monotonicity. The fuzzy monotonicity can be divided into different interval sizes to offset noise interference and improve accuracy. The calculation is mainly through the intersection and union of data and some sorting calculations, with fewer parameters. Compared with other distance measurement methods, it does not require frequent calculation of distances and more parameters. The present invention has higher classification accuracy, shorter calculation time and anti-noise interference ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a flow chart of a data mining method based on fuzzy monotonicity and entropy mentioned in Example 1 of the present invention. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0046] The present invention proposes a data mining method based on fuzzy monotone and entropy, which uses entropy values ​​to remap the relationship between continuous features and discrete decision classes, and measures the importance of features by analyzing the changes in the fuzzy monotone membership of the entropy values. In addition, the present invention also introduces a coefficient of variation to intuitively and effectively measure the changes in the fuzzy monotone membership, and parameterizes the changes in the fuzzy monotone membership values.

[0047] Compared with the prior art, the fuzzy monotonic relationship method based on inclusion degree of the present invention analyzes the fuzzy monotonic relationship by sorting continuous condition attribute features and decision attributes. It requires that both the condition and the decision attributes are continuous data. The present invention maps the fuzzy monotonic relationship between the condition attribute features and the decision class to the condition attribute features and the entropy value table through the entropy value, and performs fuzzy monotonic analysis based on the entropy value table. Compared with the former, the present invention can solve the problem of correlation analysis between the characteristics of the continuous data set of the classification task and the decision class when the condition input is continuous data and the decision output is classified data.

[0048] The fuzzy monotonic relationship method based on inclusion degree observes the change of fuzzy monotonic membership value, and by setting the post-rule, observes whether the fuzzy monotonic membership value increases continuously under the premise of satisfying greater than 0.5. The present invention describes the fluctuation of fuzzy monotonic membership value by the coefficient of variation, and sets the change of fuzzy monotonic membership value parameterized.

[0049] The feature selection methods based on neighborhood rough sets all use the distance between samples as a metric to analyze the correlation between features and decision classes. This scheme uses the change of fuzzy monotonic analysis of decision class entropy as a metric to perform correlation analysis.

[0050] The data mining method based on fuzzy monotone and entropy proposed by the present invention is described below with reference to relevant drawings and specific examples. The specific contents are as follows.

[0051] Embodiment 1:

[0052] See also Figure 1 The present invention provides a data mining method based on fuzzy monotone and entropy, comprising:

[0053] Step 1: Normalize and preprocess the original features in the data set, construct a training data set, and initialize the candidate feature reduction subset to an empty set;

[0054] Step 2, traverse the attribute features in the training data set, sort the continuous variable attribute features in ascending or descending order to obtain a sample sequence C about the decision class (the sequence C is the decision class data of the sample after the sample is sorted);

[0055] Step 3: Partition the sample sequence C. Assume that the number of elements in the interval is m, and the value range of m is [2, n / 2], where n represents the number of samples.

[0056] Step 4: Traverse the value of m and obtain:

[0057] C={C1,C2,…,C k}

[0058] Among them, k = n / m; C1 represents an interval, and the others are similar;

[0059] Step 5: Calculate each interval C k The entropy value is used to describe the degree of confusion of interval decision making, and the average value of the entropy value is further calculated to obtain the average value table of the entropy value under different k values. The specific calculation formula is:

[0060] H(X)=-∑ x p(x)logp(x)(1)

[0061]

[0062] Among them, H(X) represents the entropy value of each interval; H(D k ′) represents the average value of entropy value; p(x) represents the probability of sample x; where formula (2) represents the result of formula (1), and the average value of entropy value from interval 1 to interval k is calculated;

[0063] Step 6: The initial sequence of the entropy average value table is recorded as A, and the sequence of the entropy average value table after ascending sorting is recorded as B. Sequence A and sequence B are distinguished. Assume that the number of elements in the interval is q, and the value range of q is [2, n' / 2], where n' represents the sequence length;

[0064] Step 7: Under different q values, the fuzzy monotonic membership of the entropy value is calculated according to the fuzzy monotonic membership formula. The specific calculation formula is:

[0065]

[0066] Among them, A j represents the initial sequence in the entropy average table, C iIt represents the sorted sequence, U Av and U Cv Represents each partition of the sequence, v represents the subscript of the interval, and the mathematical symbol It means to take the conjunction of the membership values ​​of two intervals. Here, there are two fuzzy monotony cases: fuzzy monotony decreasing and fuzzy monotony increasing. In the present invention, only fuzzy monotony increasing is cited as an example. Fuzzy monotony decreasing is sorted in descending order in the interval sorting, and vice versa.

[0067] Step 8: Analyze the change of fuzzy monotone membership and observe the fluctuation of fuzzy monotone membership value through the coefficient of variation. The calculation formula of the coefficient of variation is as follows:

[0068]

[0069] Among them, fm i is a single fuzzy membership value corresponding to a certain q value; μ represents the average value of the fuzzy membership value in the entropy table;

[0070] If the coefficient of variation is less than the set threshold value θ, it indicates that the feature has a fuzzy monotonically increasing relationship with the decision class. The maximum value of the fuzzy monotonically increasing membership value is recorded, and the feature is added to the feature reduction subset.

[0071] Step 9: Check whether the features have been traversed. If not, return to S2 to continue iterating. If all features have been traversed, sort them according to the maximum value of the fuzzy monotone membership value in the entropy table of each feature, and output the reduced feature subset.

[0072] Step 10: Add the features in the feature reduction subset to the support vector machine (SVM) classifier one by one, perform iterative training, output the optimal feature sequence with the highest accuracy, and output the classification accuracy.

[0073] Embodiment 2:

[0074] Based on Example 1, but different in that, the data mining method based on fuzzy monotone and entropy proposed by the present invention is described below in conjunction with a specific example, and the specific contents are as follows:

[0075] Part of the data from the UCI wine dataset was used to conduct attribute reduction experiments. After filtering out the incomplete data and extracting the complete data, some of the data was taken out to obtain a 7*20 complete data information table 1.

[0076] Table 1

[0077] Column1 Column2 Column3 Column4 Column5 Column6 Column7 14.23 1.71 2.43 15.6 127 2.8 0 13.2 1.78 2.14 11.2 100 2.65 0 13.16 2.36 2.67 18.6 101 2.8 0 14.21 4.04 2.44 18.9 111 2.85 0 14.38 3.59 2.28 16 102 3.25 0 13.9 1.68 2.12 16 101 3.1 0 12.37 0.94 1.36 10.6 88 1.98 1 12.33 1.1 2.28 16 101 2.05 1 12.64 1.36 2.02 16.8 100 2.02 1 12.51 1.73 1.98 20.5 85 2.2 1 12.42 2.55 2.27 22 90 1.68 1 12.25 1.73 2.12 19 80 1.65 1 12.72 1.75 2.28 22.5 84 1.38 1 12.22 1.29 1.94 19 92 2.36 1 13.11 1.9 2.75 25.5 116 2.2 2 13.23 3.3 2.28 18.5 98 1.8 2 12.58 1.29 2.1 20 103 1.48 2 12.86 1.35 2.32 18 122 1.51 2 12.88 2.99 2.4 20 104 1.3 2 12.81 2.31 2.4 24 98 1.15 2

[0078] Step 1: According to the data information table, initialize the attribute set A = {a1, a2, ..., a7} and the decision set D = {0, 0, ..., 2}.

[0079] Step 2: Traverse the attributes in the attribute set. The number of traversals is the length of the attribute set. Take the first loop as an example:

[0080] Step 2.1: Take out the attribute a1 = {14.23,…,12.81}, and the corresponding decision set at this time is D = {0,0,…,2}.

[0081] Step 2.2: Sort the attribute set in ascending order, and obtain the new attribute a1 sequence a1 = {12.22,…,14.38}. The new decision set sequence is:

[0082] D={1,1,1,1,1,1,2,1,1,2,2,2,2,0,0,2,0,0,0,0}.

[0083] Step 3: Partition the sorted decision set D, and the number m in the interval is [2,10]. Here, the number of samples n of some data is 20, and n / 2 is taken to get the maximum value of m as 10. For example, when m is 4, the decision set D′4={{1,1,1,1}, {1,1,2,1}, {1,2,2,2}, {2,0,0,2}, {0,0,0,0}}, that is, D′4={D1,…,D5}, where the subscript 4 represents the number in the interval.

[0084] Step 4: According to the formula:

[0085] H(X)=-∑ x p(x)logp(x)

[0086]

[0087] Calculate the entropy value of each interval, and take the case of m=4 as an example. When m=4, calculate the entropy values ​​of the five intervals D1, D2, D3, D4, and D5. Take D2 as an example. There are two decision categories {1} and {2} in D2. Calculate the probabilities p(x) of the two decision categories respectively, and get p(1)=0.75 and p(2)=0.25. Substituting into the formula, we get H(D2)=0.81125. Similarly, we can get the entropy values ​​of the other four intervals, H(D1)=0, H(D3)=0.81125, H(D4)=1, and H(D5)=0. After averaging the entropy values ​​of the five intervals, we can get the corresponding partition entropy value H(D′4)=0.524 for m=4. Similarly, calculate the average entropy value when m takes different values ​​to get the entropy mean table:

[0088] H={0.4,0.413,0.524,0.665,0.550,0.917,0.736,0.672,0.846}.

[0089] Step 5: Record the initial sequence of the entropy mean table as X = {0,…,8}, sort the entropy mean table in ascending order, and the sorted sequence is Y = {0,1,2,4,3,7,6,8,5}

[0090] Step 6: Partition the two sequences X and Y. The number of elements p in the interval ranges from [2, n / 2], where n is the sequence length of X. In this case, the sequence length is 9. After partitioning the interval, we get Xp = {x1,…,x k}, Yp={y1,…,y k}, subscript k represents the interval number. Taking p=3 as an example, X3={{0,1,2},{3,4,5},{6,7,8}},Y3={{0,1,2},{4,3,7},{6,8,5}}, that is, X3={x1,x2,x3},Y3={y1,y2,y3}.

[0091] According to the formula:

[0092]

[0093] Calculate the inclusion degree between partitions: μ(x1,y1)=1,μ(x2,y2)=2 / 3,μ(x3,y3)=2 / 3. Conjunction of the inclusion degree is performed through the membership function formula, and μ(x3,y3)=2 / 3>0.5 is obtained.

[0094] Similarly, the membership functions of other values ​​of p are calculated to obtain the membership set F.

[0095] Step 7: For the membership set F, take the subset of the membership set length L*0.1 from the back to the front, and use the formula Calculate the coefficient of variation of the subset. If the coefficient of variation of the subset is less than the constant θ, it means that the attribute value is fuzzy monotonically increasing and is relevant to the decision class. Add attribute a1 to the candidate feature subset. If it is greater than the constant θ, return to step 2 to continue traversing the remaining attributes. It means to calculate the mean of the subset. Calculate the standard deviation of the subset. The larger the maximum value of the membership set, the higher the correlation with the decision class.

[0096] Step 8: Output the candidate feature subset to the classifier for classification.

[0097] Experimental verification:

[0098] All data of UCI's wine and musk datasets are used for experiments. The dataset description information is shown in Table 2:

[0099] Table 2

[0100]

[0101] The comparative experimental results are as follows:

[0102]

[0103] This experiment uses the classifier accuracy AC as the evaluation index and uses the rbf-SVM classifier as the classifier for algorithm verification. The higher the accuracy, the more effective the algorithm is. This algorithm achieves the best results on both the wine dataset and the musk dataset. In the classification prediction of related datasets, this algorithm can achieve the best effect.

[0104] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A data mining method based on fuzzy monotonicity and entropy, characterized in that: It includes the following: S1. Normalize and preprocess the original features in the data set, construct a training data set, and initialize the candidate feature reduction subset to an empty set; S2, traverse the attribute features in the training data set, sort the continuous variable attribute features in ascending order to obtain a sample sequence about the decision class; S3. According to the principle of interval division, the sample sequence is partitioned, the average value of the entropy values ​​of different intervals is calculated, and the interval entropy value table of the feature is obtained; S4, sorting the interval entropy value table obtained in S3 in ascending order, and calculating the fuzzy monotonic membership value of the entropy value table; S5, calculating the coefficient of variation of the fuzzy monotone membership value, if it is less than the set threshold, adding the corresponding feature to the feature reduction subset; S6, check whether the features have been traversed. If not, return to S2 to continue iterating. If all features have been traversed, sort them according to the maximum value of the fuzzy monotone membership value in the entropy table of each feature, and output the reduced feature subset. S7. Input the features in the feature reduction subset in S6 into the classifier one by one for iterative training, and output the optimal feature subset with the highest accuracy and the classification accuracy.

2. The data mining method based on fuzzy monotone and entropy according to claim 1, characterized in that: The S3 specifically includes the following contents: The sample sequence of the decision class is recorded as C, and the sample sequence C is partitioned. Assume that the number of elements in the interval is m, and the value range of m is [2, n / 2], where n represents the number of samples; traverse the value of m and obtain: C={C1,C2,…,C k } Among them, k = n / m; C1 represents an interval, and the others are similar; Calculate each interval C k The entropy value is used to describe the degree of confusion of interval decision making, and the average value of the entropy value is further calculated to obtain the average value table of the entropy value under different k values. The specific calculation formula is: H(X)=-∑ x p(x)logp(x) Among them, H(X) represents the entropy value of each interval; H(D k ′) represents the average entropy value; p(x) represents the probability of sample x.

3. The data mining method based on fuzzy monotone and entropy according to claim 2 is characterized in that: The S4 specifically includes the following contents: The initial sequence of the entropy average value table is recorded as A, and the sequence of the entropy average value table after ascending sorting is recorded as B. Sequence A and sequence B are distinguished. Assume that the number of elements in the interval is q, and the value range of q is [2,n′ / 2], where n′ represents the sequence length; Under different q values, the fuzzy monotonic membership of the entropy value is calculated according to the fuzzy monotonic membership formula. The specific calculation formula is: Among them, A j represents the initial sequence in the entropy average table, C i It represents the sorted sequence, U Av and U Cv Represents each partition of the sequence, v represents the subscript of the interval, and the mathematical symbol It means to take the conjunction of the membership values ​​of two intervals.

4. The data mining method based on fuzzy monotone and entropy according to claim 3 is characterized in that: The S5 specifically includes the following contents: Analyze the changes of fuzzy monotonic membership and observe the fluctuation of fuzzy monotonic membership value through the coefficient of variation. The calculation formula of the coefficient of variation is as follows: Among them, fm i is a single fuzzy membership value corresponding to a certain q value; μ represents the average value of the fuzzy membership value of the entropy table; It means to calculate the mean of the subset. Calculate the standard deviation of the subset; If the coefficient of variation is less than the set threshold θ, it means that the feature has a fuzzy monotonically increasing relationship with the decision class. The maximum value of the fuzzy monotonous membership value is recorded and the feature is added to the feature reduction subset.

5. The data mining method based on fuzzy monotone and entropy according to claim 4 is characterized in that: The classifier in S7 specifically refers to a support vector machine classifier.

Images