CN113610181A  Quick multitarget feature selection method combining machine learning and group intelligence algorithm  Google Patents
Quick multitarget feature selection method combining machine learning and group intelligence algorithm Download PDFInfo
 Publication number
 CN113610181A CN113610181A CN202110945987.4A CN202110945987A CN113610181A CN 113610181 A CN113610181 A CN 113610181A CN 202110945987 A CN202110945987 A CN 202110945987A CN 113610181 A CN113610181 A CN 113610181A
 Authority
 CN
 China
 Prior art keywords
 data set
 algorithm
 data
 sample data
 sample
 Prior art date
 Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
 Pending
Links
Images
Classifications

 G—PHYSICS
 G06—COMPUTING; CALCULATING OR COUNTING
 G06F—ELECTRIC DIGITAL DATA PROCESSING
 G06F18/00—Pattern recognition
 G06F18/20—Analysing
 G06F18/23—Clustering techniques
 G06F18/232—Nonhierarchical techniques
 G06F18/2321—Nonhierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
 G06F18/23213—Nonhierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. Kmeans clustering

 G—PHYSICS
 G06—COMPUTING; CALCULATING OR COUNTING
 G06F—ELECTRIC DIGITAL DATA PROCESSING
 G06F18/00—Pattern recognition
 G06F18/20—Analysing
 G06F18/24—Classification techniques

 G—PHYSICS
 G06—COMPUTING; CALCULATING OR COUNTING
 G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
 G06N20/00—Machine learning
Landscapes
 Engineering & Computer Science (AREA)
 Data Mining & Analysis (AREA)
 Theoretical Computer Science (AREA)
 Physics & Mathematics (AREA)
 Evolutionary Computation (AREA)
 Artificial Intelligence (AREA)
 Computer Vision & Pattern Recognition (AREA)
 General Physics & Mathematics (AREA)
 General Engineering & Computer Science (AREA)
 Bioinformatics & Cheminformatics (AREA)
 Bioinformatics & Computational Biology (AREA)
 Evolutionary Biology (AREA)
 Life Sciences & Earth Sciences (AREA)
 Software Systems (AREA)
 Probability & Statistics with Applications (AREA)
 Medical Informatics (AREA)
 Computing Systems (AREA)
 Mathematical Physics (AREA)
 Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
A quick multitarget feature selection method combining machine learning and group intelligence algorithm comprises the following steps: 1) preprocessing a data set to be processed, 2) selecting sample data with large influence on classification performance from the processed data set based on a Kmeans clustering algorithm in machine learning, 3) providing a utilization strategy of the rest data sets, 4) processing a multitarget feature selection problem by using a group intelligent algorithm, and outputting the obtained Pareto optimal solution set and the operation time of the algorithm. According to the invention, the sample data with large influence on the classification performance is selected by the Kmeans clustering algorithm, so that the time cost of the group intelligent algorithm when the KNN classifier is called to evaluate the individual quality is reduced. And the full training of the KNN classifier is ensured by the residual sample utilization strategy. By combining the feature selection and sample selection, the time cost for solving the feature subset is reduced on the basis of ensuring the quality of the feature subset obtained by the swarm intelligence algorithm.
Description
Technical Field
The invention belongs to the field of intelligent optimization and feature selection, and relates to a quick multitarget feature selection method combining machine learning and a group intelligent algorithm.
Background
In this explosive data era, real data generated by thousands of industries contains a great deal of redundancy and features with complex interaction relationships, which seriously affect the performance of data analysis methods. And too high a data dimension can lead to a sharp increase in the time complexity of the algorithm processing. The feature selection is a common data dimension reduction method, and the main purpose of the method is to select important features from the original feature set to form a new feature subset on the basis of not sacrificing the learning performance of the original feature set, so that the calculation efficiency of the algorithm is accelerated, and the generalization performance of the model is enhanced. The feature selection method is commonly used in machine learning tasks such as classification and clustering, so that the quality of the feature set is improved.
Traditional optimization methods based on gradient or direct search are difficult to apply since the feature selection problem is usually largescale, multimodal, nonlinear and the objective function has no clear functional form. In addition, in real production life, the acquisition of each feature data is costly, so that in addition to paying attention to the classification accuracy of the selected feature subsets, the number of features included in the selected feature subsets needs to be considered, that is, feature selection is regarded as a multiobjective optimization problem. Because the group intelligent algorithm has strong global searching capability, the mechanism based on the group can generate a plurality of solutions in one operation. Therefore, the group intelligence algorithm is very suitable for the multitarget feature selection problem, and can carry out balance between the feature quantity and the classification performance so as to find a group of Pareto dominant solutions. At present, common swarm intelligence algorithms such as a particle swarm algorithm and an artificial bee swarm algorithm are applied to the problem of multitarget feature selection.
However, when the group intelligent algorithm is used for processing the feature selection problem, the method depends on multiple iterations of the group, and each iteration needs to classify the highdimensional data to be processed, so that the advantages and disadvantages of the feature subset selected by the current algorithm can be evaluated according to the classification result and the number of the selected features at the moment, and the group evolution is guided. Therefore, group intelligence algorithms typically have a high computational cost when dealing with feature selection problems. For many practical engineering problems, the timeliness of the operation of the algorithm determines the practical application value of the algorithm, so that how to reduce the operation time cost of the group intelligent algorithm on the basis of ensuring that the solving quality of the group intelligent algorithm is not sacrificed is an urgent problem to be solved.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention provides a quick multitarget feature selection method combining machine learning and a group intelligent algorithm. The method combines the problem of feature selection and sample selection, and reduces the time cost of solving the problem on the basis of ensuring the quality of a feature subset obtained by a group intelligent algorithm.
The technical scheme adopted by the invention is as follows:
a quick multitarget feature selection method combining machine learning and group intelligence algorithm comprises the following steps:
(a1) each row of data in the data set U' is called sample data; mapping the classification label corresponding to each sample data in the data set U 'to a natural number set and adjusting the classification label to a first position in the corresponding sample data, then rearranging the sample data according to the size relation of the classification labels of the sample data from small to big, and marking the adjusted data set as U';
(b1) carrying out normalization processing on the data set U' to obtain a data set U; by U_{i}′_{j}And U_{ij}Data of ith row and j column in the data sets U' and U respectively are represented, wherein i belongs to {1, 2.., m }, and j belongs to {2, 3.., n };andmaximum and minimum values in column j of data set U', respectively, then U_{ij}The formula (1) can be used for calculating:
step 2: based on machine learningThe Kmeans clustering algorithm selects sample data with large influence on classification performance from the data set U to form an important data set D_{import}The remaining samples constitute the data set D_{other}The method comprises the following specific steps:
(a2) analyzing a first column of data of a data set U, determining that sample data has K types, setting that the number of clustering centers of a Kmeans clustering algorithm is K, and expressing the obtained pth clustering center by p, wherein p belongs to {1, 2.. multidot.K }, and initializing p to be 1;
(b2) clustering the data set U according to a Kmeans method, adding the clustering label of each sample to the front of the classification label of each sample, and rearranging the sample data according to the sequence of the clustering labels from small to large to form a new data set
(c2) Storing sample data with a clustering label p into a class (p), and performing comparative analysis on the clustering label and the classification label on the sample data in the class (p): first, find the most numerous class labels and mark them as label_{max}And all classes are labeled label_{max}Put the sample data of into the matrix U_{same}(p) in (c); by labelling classes other than label_{max}Put the sample data of into the matrix U_{diff}(p) in (c);
(d2) handle U_{diff}(p) all sample data in the set is added to the data setBecause these elements have a very important role in the correct classification; at the same time from U_{same}(p) randomly selecting a small number of samples to add to the data setBecause of U_{same}The sample data in (p) has high similarity, and a small number of samples can be randomly selected to well replace the sample data, U_{same}(p) putting the remaining sample data in the data set
(e2) Let p ≠ p +1, if p ≠ K, then step (c2) is skipped; otherwise will respectivelyp is formed by combining {1, 2.. multidata.,. K }Data set ofp is formed by combining {1, 2.. multidata.,. K }A data set;
(f2) removing data setsAndthe cluster label in (1) is D, the data set composed of sample data with great influence on the classification performance is obtained_{import}The data set composed of the remaining samples is D_{other}；
And step 3: through step 2 pick out D_{import}The data set is utilized in the early stage of the operation of the group intelligent algorithm to reduce the time cost of the algorithm for evaluating individuals by calling the classifier; however, the classifier training may be insufficient due to the fact that fewer samples are used, and in order to ensure the accuracy of the solution result, the data set D needs to be determined_{other}I.e. gradually turning D in the later stage of the group intelligence algorithm operation_{other}The sample data in the method are gradually added into a training set of the classifier, so that the swarm intelligence algorithm can obtain a highquality feature subset finally; the method comprises the following specific steps:
(a3) setting the maximum iteration number of the group intelligent algorithm as iter, wherein t represents the operation of the algorithm for the t time, and belongs to {1, 2., iter }; before the algorithm, iter_{1}In the subiteration operation, only D is added_{import}The data set being used to evaluate solutions generated by the algorithmQuality;
(b3) definition of size (U) and size (D)_{import}) Representing data sets U and D, respectively_{import}The number of the sample data in the middle stage, and further constructing a threshold function by using a Sigmoid function as follows:
(c3) according to the threshold function (2), a data set D is given_{other}The specific utilization strategy of (1) is as follows: after iteriter of group intelligence algorithm_{1}In a second iteration run, from D_{other}Randomly selecting a plurality of samples in the data set to be supplemented into a training set of the classifier, so that the total sample data of the training set is addfunc (t);
and 4, step 4: based on the two data sets D obtained in step 2_{import}And D_{other}And using the data set D given in step 3_{other}The multitarget feature selection problem is processed by using a group intelligent algorithm, and a Pareto optimal solution set obtained by the algorithm and the operation time of the algorithm are output; the method comprises the following specific steps:
(a4) initializing a group intelligence algorithm: setting operation parameters such as the population scale of the algorithm, the number of targets to be optimized, the maximum iteration times of the algorithm, the algorithm termination condition and the like; setting a constraint condition of each individual in the population, and randomly generating a plurality of individuals under the constraint condition to form an initial population; each individual is a vector subjected to coding processing and represents a feasible solution;
(b4) introduction of D produced in step 2_{import}And D_{other}A data set;
(c4) judging whether the iteration running times t is less than iter_{1}If yes, the data set D is processed_{import}As a data set for training the KNN classifier, the data set D given in step 3 is used if not_{other}Generating a dynamic data set D 'by using the strategy, and taking the D' as a data set for training the KNN classifier;
(d4) evaluating the advantages and the disadvantages: firstly, each individual in the population is processed by utilizing a proper decoding strategy, and the current individual represents which characteristics in the data set are selected through decoding analysis; two targets to be optimized, target 1(Count), are then determined: statistics of the number of features selected, target 2 (Error): calling a KNN classifier commonly used in machine learning to calculate classification accuracy; evaluating the advantages and disadvantages of individuals in the current population based on a Pareto domination relationship, and guiding the evolution direction of a group intelligent algorithm according to an evaluation result;
(e4) the group intelligence algorithm generates new individuals: opening a running timer of a program, and guiding the evolution direction of the group intelligent algorithm by using the quality results of the individuals in the group obtained based on the Pareto domination relationship in the step (d4), namely the group intelligent algorithm executes a specific search evolution mechanism according to the evaluation result to generate new individuals;
(f4) judging whether a final value condition of the algorithm is met, if not, jumping to the step (c4) for execution; and if so, stopping the running timer of the program, and outputting the running time of the algorithm and the Pareto optimal solution set obtained by the algorithm aiming at the multitarget feature selection problem.
Compared with the prior art, the method selects the sample data with large influence on the classification performance through the Kmeans clustering algorithm, thereby reducing the time cost of the group intelligent algorithm when the KNN classifier is called to evaluate the individual quality. And the full training of the KNN classifier is ensured by the residual sample utilization strategy, so that the accuracy of the solving result is ensured. By combining the feature selection and sample selection, the time cost for solving the feature subset is reduced on the basis of ensuring the quality of the feature subset obtained by the swarm intelligence algorithm.
Drawings
FIG. 1 is a flow chart of the steps of the present invention.
FIG. 2 is a schematic diagram of the Kmeans algorithm used to sort out sample data from a data set that has a large impact on the classification performance.
FIG. 3 is a flow chart of a group intelligence algorithm processing the multiobjective feature selection problem based on the policies set forth in step 2 and step 3.
Detailed Description
The invention is further described below with reference to the accompanying drawings. As shown in fig. 1, a fast multitarget feature selection method combining machine learning and group intelligence algorithm includes the following steps:
(a1) each row of data in the data set U' is referred to as a sample data. And mapping the classification label corresponding to each sample data in the data set U 'to a natural number set and adjusting the classification label to the first position in the corresponding sample data, rearranging the sample data according to the size relation of the classification labels of the sample data from small to large, and marking the adjusted data set as U'.
(b1) And carrying out normalization processing on the data set U' to obtain a data set U. By U_{i}′_{j}And U_{ij}Data of ith row and j column in data sets U' and U respectively are represented, wherein i belongs to {1, 2.Andmaximum and minimum values in column j of data set U', respectively, then U_{ij}The formula (1) can be used for calculating:
step 2: based on a Kmeans clustering algorithm in machine learning, sample data which has great influence on classification performance is selected from a data set U to form an important data set D_{import}The remaining samples constitute the data set D_{other}The method comprises the following specific steps:
(a2) analyzing the first column of data of the data set U, determining that the sample data has K types, setting that the number of clustering centers of a Kmeans clustering algorithm is K, and expressing the obtained pth clustering center by using p, wherein p belongs to {1, 2. In fig. 2K is 3.
(b2) Clustering the data set U according to a Kmeans method, adding the clustering label of each sample to the front of the classification label of each sample, and rearranging the sample data according to the sequence of the clustering labels from small to large to form a new data setThis process is illustrated in fig. 2.
(c2) Storing sample data with a clustering label p into a class (p), and performing comparative analysis on the clustering label and the classification label on the sample data in the class (p): first, find the most numerous class labels and mark them as label_{max}And all classes are labeled label_{max}Put the sample data of into the matrix U_{same}(p) in (c). By labelling classes other than label_{max}Put the sample data of into the matrix U_{diff}(p) in (c). In part (b) of fig. 2, the sample data enclosed by the dotted line is U_{diff}(p) remaining data not framed is U_{same}(p)。
(d2) Handle U_{diff}(p) all sample data in the set is added to the data setSince these elements have a very important role in the correct classification. At the same time from U_{same}(p) randomly selecting a small number of samples to add to the data setBecause of U_{same}The sample data in (p) has high similarity, and a small number of samples can be randomly selected to well replace the sample data, U_{same}(p) putting the remaining sample data in the data set
(e2) Let p be p +1, and if p ≠ K, go to step (c 2). Otherwise will respectivelyp is formed by combining {1, 2.. multidata.,. K }Data set ofp is formed by combining {1, 2.. multidata.,. K }A data set.
(f2) Removing data setsAndthe cluster label in (1) is D, the data set composed of sample data with great influence on the classification performance is obtained_{import}The data set composed of the remaining samples is D_{other}。
And step 3: through step 2 pick out D_{import}A data set that is utilized in a prestage of the group intelligence algorithm's operation to reduce the time cost of the algorithm in invoking the classifier to evaluate the individual. However, the classifier training may be insufficient due to the fact that fewer samples are used, and in order to ensure the accuracy of the solution result, the data set D needs to be determined_{other}I.e. gradually turning D in the later stage of the group intelligence algorithm operation_{other}The sample data in the method is gradually added into a training set of the classifier, so that the swarm intelligence algorithm can obtain a highquality feature subset finally. The method comprises the following specific steps:
(a3) and setting the maximum iteration number of the group intelligent algorithm as iter, wherein t represents the tth operation of the algorithm, and belongs to {1, 2., iter }. Before the algorithm, iter_{1}In the subiteration operation, only D is added_{import}The data set is used to assess the quality of the solution produced by the algorithm.
(b3) Definition of size (U) and size (D)_{import}) Representing data sets U and D, respectively_{import}The number of the sample data in the middle stage, and further constructing a threshold function by using a Sigmoid function as follows:
(c3) according to the threshold function (2), a data set D is given_{other}The specific utilization strategy of (1) is as follows: after iteriter of group intelligence algorithm_{1}In a second iteration run, from D_{other}Randomly selecting a plurality of samples in the data set to be supplemented into the training set of the classifier, so that the total sample data of the training set is addfunc (t).
And 4, step 4: based on the two data sets D obtained in step 2_{import}And D_{other}And using the data set D given in step 3_{other}The multitarget feature selection problem is processed by using a group intelligent algorithm, and a Pareto optimal solution set obtained by the algorithm and the operation time of the algorithm are output. Fig. 3 shows a flowchart of step 4, which includes the following steps:
(a4) initializing a group intelligence algorithm: and setting operation parameters of the algorithm, such as population scale, the number of targets to be optimized, the maximum iteration times of the algorithm, algorithm termination conditions and the like. And setting a constraint condition of each individual in the population, and randomly generating a plurality of individuals under the constraint condition to form an initial population. Each of the abovementioned individuals is a vector subjected to encoding processing, and represents a feasible solution.
(b4) Introduction of D produced in step 2_{import}And D_{other}A data set.
(c4) Judging whether the iteration running times t is less than iter_{1}If yes, the data set D is processed_{import}As a data set for training the KNN classifier, the data set D given in step 3 is used if not_{other}And (3) generating a dynamic data set D 'by using the strategy, wherein the D' is used as a data set for training the KNN classifier.
(d4) Evaluating the advantages and the disadvantages: firstly, each individual in the population is processed by utilizing a proper decoding strategy, and the current individual represents which characteristics in the data set are selected through decoding analysis; two targets to be optimized, target 1(Count), are then determined: statistics of the number of features selected, target 2 (Error): calling a KNN classifier commonly used in machine learning to calculate classification accuracy; evaluating the advantages and disadvantages of individuals in the current population based on a Pareto domination relationship, and guiding the evolution direction of a group intelligent algorithm according to an evaluation result;
(e4) the group intelligence algorithm generates new individuals: and (d) starting a running timer of the program, and guiding the evolution direction of the group intelligent algorithm by using the quality results of the individuals in the group obtained based on the Pareto domination relationship in the step (d4), namely, the group intelligent algorithm executes a specific search evolution mechanism according to the evaluation result to generate new individuals.
(f4) Judging whether a final value condition of the algorithm is met, if not, jumping to the step (c4) for execution; and if so, stopping the running timer of the program, and outputting the running time of the algorithm and the Pareto optimal solution set obtained by the algorithm aiming at the multitarget feature selection problem.
Claims (1)
1. A quick multitarget feature selection method combining machine learning and group intelligent algorithm is characterized by comprising the following steps:
(a1) each row of data in the data set U' is called sample data; mapping the classification label corresponding to each sample data in the data set U 'to a natural number set and adjusting the classification label to a first position in the corresponding sample data, rearranging the sample data according to the size relation of the classification labels of the sample data from small to big, and marking the adjusted data set as U';
(b1) carrying out normalization processing on the data set U' to obtain a data set U; from U'_{ij}And U_{ij}Data of ith row and j column in the data sets U' and U respectively are represented, wherein i belongs to {1, 2.., m }, and j belongs to {2, 3.., n };andmaximum and minimum values in column j of data set U', respectively, then U_{ij}The formula (1) can be used for calculating:
step 2: based on a Kmeans clustering algorithm in machine learning, sample data which has great influence on classification performance is selected from a data set U to form an important data set D_{import}The remaining samples constitute the data set D_{other}The method comprises the following specific steps:
(a2) analyzing a first column of data of a data set U, determining that sample data has K types, setting that the number of clustering centers of a Kmeans clustering algorithm is K, and expressing the obtained pth clustering center by p, wherein p belongs to {1, 2.. multidot.K }, and initializing p to be 1;
(b2) clustering the data set U according to a Kmeans method, adding the clustering label of each sample to the front of the classification label of each sample, and rearranging the sample data according to the sequence of the clustering labels from small to large to form a new data set
(c2) Storing sample data with a clustering label p into a class (p), and performing comparative analysis on the clustering label and the classification label on the sample data in the class (p): first, find the most numerous class labels and mark them as label_{max}And all classes are labeled label_{max}Put the sample data of into the matrix U_{same}(p) in (c); by labelling classes other than label_{max}Put the sample data of into the matrix U_{diff}(p) in (c);
(d2) handle U_{diff}(p) all sample data in the set is added to the data setBecause these elements have a very important role in the correct classification; at the same time from U_{same}(p) randomly selecting a small number of samplesJoining to a data setBecause of U_{same}The sample data in (p) has high similarity, and a small number of samples can be randomly selected to well replace the sample data, U_{same}(p) putting the remaining sample data in the data set
(e2) Let p ≠ p +1, if p ≠ K, then step (c2) is skipped; otherwise will respectivelyAre combined intoData set ofAre combined intoA data set;
(f2) removing data setsAndthe cluster label in (1) is D, the data set composed of sample data with great influence on the classification performance is obtained_{import}The data set composed of the remaining samples is D_{other}；
And step 3: through step 2 pick out D_{import}The data set is utilized in the early stage of the operation of the group intelligent algorithm to reduce the time cost of the algorithm for evaluating individuals by calling the classifier; to ensure the accuracy of the solution result, the data set D needs to be determined_{other}By means ofSlightly, i.e. gradually moving D in the later stages of the group intelligence algorithm run_{other}The sample data in the method are gradually added into a training set of the classifier, so that the swarm intelligence algorithm can obtain a highquality feature subset finally; the method comprises the following specific steps:
(a3) setting the maximum iteration number of the group intelligent algorithm as iter, wherein t represents the operation of the algorithm at the t time, and belongs to {1, 2.. i.t, er; before the algorithm, iter_{1}In the subiteration operation, only D is added_{import}The data set is used to evaluate the quality of the solution generated by the algorithm;
(b3) definition of size (U) and size (D)_{import}) Representing data sets U and D, respectively_{import}The number of the sample data in the middle stage, and further constructing a threshold function by using a Sigmoid function as follows:
(c3) according to the threshold function (2), a data set D is given_{other}The specific utilization strategy of (1) is as follows: after iteriter of group intelligence algorithm_{1}In a second iteration run, from D_{other}Randomly selecting a plurality of samples in the data set to be supplemented into a training set of the classifier, so that the total sample data of the training set is addfunc (t);
and 4, step 4: based on the two data sets D obtained in step 2_{import}And D_{other}And using the data set D given in step 3_{other}The multitarget feature selection problem is processed by using a group intelligent algorithm, and a Pareto optimal solution set obtained by the algorithm and the operation time of the algorithm are output; the method comprises the following specific steps:
(a4) initializing a group intelligence algorithm: setting operation parameters such as the population scale of the algorithm, the number of targets to be optimized, the maximum iteration times of the algorithm, the algorithm termination condition and the like; setting a constraint condition of each individual in the population, and randomly generating a plurality of individuals under the constraint condition to form an initial population; each individual is a vector subjected to coding processing and represents a feasible solution;
(b4) introduction of D produced in step 2_{import}And D_{other}A data set;
(c4) judging whether the iteration running times t is less than iter_{1}If yes, the data set D is processed_{import}As a data set for training the KNN classifier, the data set D given in step 3 is used if not_{other}Generating a dynamic data set D 'by using the strategy, and taking the D' as a data set for training the KNN classifier;
(d4) evaluating the advantages and the disadvantages: firstly, each individual in the population is processed by utilizing a proper decoding strategy, and the current individual represents which characteristics in the data set are selected through decoding analysis; two targets to be optimized, target 1(Count), are then determined: statistics of the number of features selected, target 2 (Error): calling a KNN classifier commonly used in machine learning to calculate classification accuracy; evaluating the advantages and disadvantages of individuals in the current population based on a Pareto domination relationship, and guiding the evolution direction of a group intelligent algorithm according to an evaluation result;
(e4) the group intelligence algorithm generates new individuals: opening a running timer of a program, and guiding the evolution direction of the group intelligent algorithm by using the quality results of the individuals in the group obtained based on the Pareto domination relationship in the step (d4), namely the group intelligent algorithm executes a specific search evolution mechanism according to the evaluation result to generate new individuals;
(f4) judging whether a final value condition of the algorithm is met, if not, jumping to the step (c4) for execution; and if so, stopping the running timer of the program, and outputting the running time of the algorithm and the Pareto optimal solution set obtained by the algorithm aiming at the multitarget feature selection problem.
Priority Applications (1)
Application Number  Priority Date  Filing Date  Title 

CN202110945987.4A CN113610181A (en)  20210817  20210817  Quick multitarget feature selection method combining machine learning and group intelligence algorithm 
Applications Claiming Priority (1)
Application Number  Priority Date  Filing Date  Title 

CN202110945987.4A CN113610181A (en)  20210817  20210817  Quick multitarget feature selection method combining machine learning and group intelligence algorithm 
Publications (1)
Publication Number  Publication Date 

CN113610181A true CN113610181A (en)  20211105 
Family
ID=78341041
Family Applications (1)
Application Number  Title  Priority Date  Filing Date 

CN202110945987.4A Pending CN113610181A (en)  20210817  20210817  Quick multitarget feature selection method combining machine learning and group intelligence algorithm 
Country Status (1)
Country  Link 

CN (1)  CN113610181A (en) 
Cited By (1)
Publication number  Priority date  Publication date  Assignee  Title 

CN115861672B (en) *  20221220  20230919  中南大学  Foam flotation operation performance evaluation method based on image feature joint distribution 

2021
 20210817 CN CN202110945987.4A patent/CN113610181A/en active Pending
Cited By (1)
Publication number  Priority date  Publication date  Assignee  Title 

CN115861672B (en) *  20221220  20230919  中南大学  Foam flotation operation performance evaluation method based on image feature joint distribution 
Similar Documents
Publication  Publication Date  Title 

Jiang et al.  Asymmetric deep supervised hashing  
Donahue et al.  Decaf: A deep convolutional activation feature for generic visual recognition  
CN112132179A (en)  Incremental learning method and system based on small number of labeled samples  
CN112069310B (en)  Text classification method and system based on active learning strategy  
Yan et al.  Unsupervised and semi‐supervised learning: The next frontier in machine learning for plant systems biology  
US11908457B2 (en)  Orthogonally constrained multihead attention for speech tasks  
CN103258210B (en)  A kind of highdefinition image classification method based on dictionary learning  
CN110188827B (en)  Scene recognition method based on convolutional neural network and recursive automatic encoder model  
Chen et al.  Sparse spatial transformers for fewshot learning  
CN111325264A (en)  Multilabel data classification method based on entropy  
CN113541834B (en)  Abnormal signal semisupervised classification method and system and data processing terminal  
WO2018224165A1 (en)  Device and method for clustering a set of test objects  
CN117150026B (en)  Text content multilabel classification method and device  
CN116594748A (en)  Model customization processing method, device, equipment and medium for task  
CN115037805A (en)  Unknown network protocol identification method, system, device and storage medium based on deep clustering  
CN115795065A (en)  Multimedia data crossmodal retrieval method and system based on weighted hash code  
Chouaib et al.  Feature selection combining genetic algorithm and adaboost classifiers  
CN111191033A (en)  Open set classification method based on classification utility  
US20220101133A1 (en)  Dynamic quantization for energy efficient deep learning  
CN113610181A (en)  Quick multitarget feature selection method combining machine learning and group intelligence algorithm  
Fonseca et al.  Modelagnostic approaches to handling noisy labels when training sound event classifiers  
CN111783688B (en)  Remote sensing image scene classification method based on convolutional neural network  
CN112990371A (en)  Unsupervised night image classification method based on feature amplification  
CN115758462A (en)  Method, device, processor and computer readable storage medium for realizing sensitive data identification in trusted environment  
Du et al.  Deep neural networks with parallel autoencoders for learning pairwise relations: Handwritten digits subtraction 
Legal Events
Date  Code  Title  Description 

PB01  Publication  
PB01  Publication  
SE01  Entry into force of request for substantive examination  
SE01  Entry into force of request for substantive examination 