Feature selection method and device for predicting inhalation allergy of children

Abstract

The invention discloses a feature selection method and device for children inhalation allergy prediction, and relates to the technical field of data processing.The method comprises the steps that firstly, a population is searched through a gradient search optimization algorithm, so that a reasonable and targeted local search direction can be formed near pathological features and clinical manifestation features; rapid convergence is carried out on an area with better prediction performance, and the search efficiency is improved; the current solution can be moderately and randomly expanded in the feature space in the search process, so that the exploration range of the algorithm in the high-dimensional space formed by numerous clinical and pathological features is effectively expanded, and the search process is prevented from being caught in local optimum too early. And fine tuning is carried out on the current optimal feature subset through chaotic local search, so that the convergence speed is accelerated on the basis of jumping out of local optimum, the calculation efficiency is considered while relatively high prediction precision is ensured, and key features are identified for subsequent use.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a feature selection method, device, equipment and medium for predicting inhaled allergy of children. BACKGROUND

[0002] Childhood allergic diseases, including atopic dermatitis, asthma, allergic rhinitis and food allergy, are increasingly prevalent chronic diseases worldwide, causing serious impact on children's health and quality of life. Inhaled allergy, caused by airborne allergens such as dust mites, pollen, mold and animal dander, is a common type of allergy and the main cause of respiratory allergic diseases such as asthma and allergic rhinitis. Therefore, its early identification and intervention are of great significance. With the rapid development of computer science and artificial intelligence technology, machine learning has been widely applied in the prediction of allergic diseases. Through the analysis of a large amount of clinical and environmental data, machine learning models can automatically identify potential risk factors and sensitization pathways, thereby achieving early prediction and individualized management of inhaled allergy in children. In the process of building an inhaled allergy prediction model, due to the problems of high dimensionality, feature redundancy and noise interference of clinical and environmental data, if all variables are directly used for modeling, it will often lead to increased model complexity and decreased prediction performance. Therefore, it is necessary to select key features that have a significant impact on the prediction results from the original data through feature selection methods, and eliminate noise or irrelevant variables, so as to improve the prediction accuracy and stability of the model.

[0003] Existing feature selection methods can be generally divided into three categories: filter-based, wrapper-based and embedded-based methods. Filter-based methods evaluate the correlation between each feature and the target variable through statistical tests, regardless of the specific learning algorithm. Embedded methods combine regularization techniques to achieve feature selection during model training. Wrapper methods directly use model performance as an evaluation standard, and evaluate the pros and cons of feature subsets through their actual prediction results. However, since wrapper methods need to search in a large number of feature combinations, it is extremely difficult to exhaust all possible feature subsets in terms of computation, especially in high-dimensional data situations.

[0004] In recent years, meta-heuristic algorithms have been widely applied to feature selection tasks due to their excellent global search ability and adaptability. By simulating the optimization mechanisms of group cooperation or biological behavior in nature, meta-heuristic algorithms can effectively balance global exploration and local development in complex search spaces, thereby finding better feature subsets. However, feature selection methods based on meta-heuristic algorithms are prone to fall into local optima in the later stages of search, and the convergence speed of the algorithm is slow, resulting in unstable feature selection results. Therefore, at present, feature selection methods based on meta-heuristic algorithms are difficult to balance high prediction accuracy and computational efficiency, ultimately making it difficult to effectively identify key features. SUMMARY

[0005] The embodiment of the present application provides a feature selection method and device for predicting inhaled allergy of children, and can solve the problems in the prior art.

[0006] The embodiment of the present application provides a feature selection method for predicting inhaled allergy of children, comprising the following steps: obtaining a feature data set associated with inhaled allergy of a child patient, the feature data set comprising pathological features and clinical features; initializing a population based on the feature data set, determining a local search direction of the pathological features and the clinical features associated with the inhaled allergy of the child by using a gradient search optimization algorithm, guiding individuals in the initialized population to perform global search in a feature space based on the local search direction, obtaining a global optimal feature subset, two candidate solutions and two random solutions, and combining the global optimal feature subset, the two candidate solutions and the two random solutions by using a local escape operator to obtain a plurality of new feature subsets; determining an expansion direction of interaction between the pathological features and the clinical features associated with the inhaled allergy of the child by using a dispersed foraging search algorithm, guiding the plurality of new feature subsets to perform random expansion in the feature space according to the expansion direction to obtain a plurality of interactive feature subsets, and guiding the plurality of interactive feature subsets to perform local disturbance search in the feature space by using a chaotic local search algorithm to obtain a plurality of final feature subsets; performing fitness evaluation on the plurality of final feature subsets, and repeating the search until a maximum iteration number is reached to obtain an optimal feature subset.

[0007] Preferably, the search of the gradient search optimization algorithm comprises: The population is represented as wherein the population size is each individual represents a coding vector of a candidate feature subset scheme, and the dimension of the vector is the total number of features contained in the candidate feature data set; When the gradient search optimization algorithm searches, gradient search rules GSR are used to search, the GSR uses movement direction DM information of the individual in the feature space, judges the influence trend of feature addition or deletion on the prediction accuracy according to the performance feedback of the current feature subset on the inhaled allergy prediction model, and then the potential position update formula of the current individual is represented as: ; ; ; ; wherein: and is a random number in the interval [0, 1]; is the current best feature subset; and denote the value of the th candidate feature subset and the current best feature subset in the th dimension, respectively; is a new vector generated based on GSR and DM; parameter is used to balance the global exploration and local exploitation stages of the algorithm; ; ; ; ; wherein: is a random number following the standard normal distribution; is a decimal number in the interval [0, 1]; ; ; ; ; ; ; and denote the best solution and the worst solution in the population, respectively; is a random integer selected from the interval [1, N] and different from each other; ; ; ; ; The mathematical definition of the motion direction term DM is represented as: ; ; ; wherein: and have the same definition, is a random number in the interval [0, 1].

[0008] Preferably, the search process of the local escape operator comprises: After the gradient search optimization algorithm searches, the local escape operator generates a new feature subset solution by combining the current global optimal feature subset , two candidate solutions with good predictive performance , and two random solutions ; the mathematical expression is:​ ; ; ; ; wherein: is a uniform random number within the interval [0, 1], obeys a normal distribution with mean 0 and standard deviation 1; and is a random number within the interval [0, 1]; the parameter is defined as: ; ; ; wherein: is a binary parameter with value 0 or 1; when , ; otherwise ; is a new solution generated within the domain of the feature space, denoted as: ; ; wherein: the definition of is the same as is a randomly selected set of feature subset solutions in the population; and are the lower and upper bounds of the search space, respectively; denotes the dimension of the feature space.

[0009] Preferably, the search process of the scatter foraging search comprises: the scatter foraging search simulates the scattered and random foraging behavior of individuals in nature, randomly expands the current solution in the feature space, so as to enhance the search range and increase the candidate feature combination; the scatter foraging search is denoted as: ; ; ; wherein: and denote the positions of the th candidate feature subset in the current iteration and the next iteration, respectively. and denote two different individuals randomly selected from the population, and satisfy ; the parameter is a step control parameter, which is subject to a normal distribution with a mean of 0.5 and a standard deviation of 0.1; the adaptive diffusion rate DR is used to control the randomness of the search, and the value range is ; denotes the current number of function evaluations; denotes the maximum number of function evaluations.

[0010] Preferably, the search process of the chaotic local search includes: The chaotic local search performs local perturbation and fine-tuning on the pathological features and clinical indicators related to inhalant allergy that have been preliminarily screened out; The chaotic local search generates a chaotic sequence using a Logistic chaotic mapping, which is represented as: ; wherein: , and ; when the control parameter , the Logistic function enters a fully chaotic state, thereby being able to generate a uniformly distributed chaotic sequence; Based on the Logistic chaotic mapping, the chaotic local search function is represented as: ; wherein: and are the lower and upper bounds of the search space, respectively; the chaotic vector is mapped from the chaotic variable ; Using the chaotic vector and the current optimal solution , the formula for generating a new candidate feature subset is: ; ; wherein: denotes the new candidate feature subset.

[0011] Preferably, after the feature data set is obtained, the feature data set is further subjected to feature filtering, including: The feature filtering of the feature data set is performed by mutual information, Relief, information gain, and F-Score to filter the pathological features and clinical features; The mutual information is used to evaluate the correlation strength between each type of clinical and environmental feature and the presence of inhalant allergy; The Relief evaluates the importance of various clinical and environmental characteristics by measuring the difference between the presence of inhaled allergy samples and the absence of inhaled allergy samples; The information gain is based on the concept of entropy and is used to measure the degree of uncertainty reduction of data classification given a certain clinical or environmental characteristic; The F-Score is used to evaluate the importance of various clinical and environmental characteristics by measuring the difference between the presence of inhaled allergy samples and the absence of inhaled allergy samples.

[0012] The embodiment of the present application also provides a feature selection device for predicting inhaled allergy of children, which comprises: A data acquisition module is configured to acquire a feature data set associated with inhaled allergy of children, wherein the feature data set comprises pathological characteristics and clinical characteristics; A feature selection module is configured to initialize a population based on the feature data set, determine a local search direction of the pathological characteristics and the clinical characteristics associated with inhaled allergy of children by using a gradient search optimization algorithm, guide individuals in the initialized population to perform global search in a feature space based on the local search direction, obtain a global optimal feature subset, two candidate solutions and two random solutions, and obtain a plurality of new feature subsets by combining the global optimal feature subset, the two candidate solutions and the two random solutions by using a local escape operator. An extended direction of interaction between the pathological characteristics and the clinical characteristics associated with inhaled allergy of children is determined by using a dispersed foraging search algorithm, and the plurality of new feature subsets are guided to perform random expansion in the feature space according to the extended direction, so as to obtain a plurality of interactive feature subsets; and the plurality of interactive feature subsets are guided to perform local disturbance search in the feature space by using a chaotic local search algorithm, so as to obtain a plurality of final feature subsets. The plurality of final feature subsets are subjected to fitness evaluation, and the search is repeated until a maximum iteration number is reached, so as to obtain an optimal feature subset.

[0013] The embodiment of the present application also provides an electronic device, which comprises a memory and a processor. The memory is configured to store a computer program. The processor is configured to execute the computer program stored in the memory, so as to realize the steps of the feature selection method for predicting inhaled allergy of children.

[0014] The embodiment of the present application also provides a computer readable storage medium, which is configured to store a computer program, and the computer program is executed by a processor to realize the steps of the feature selection method for predicting inhaled allergy of children.

[0015] The embodiment of the present application provides a feature selection method and device for predicting inhaled allergy of children, and has the following beneficial effects compared with the prior art. The present application firstly performs importance filtering on the feature data set to form a candidate feature data set; after initializing the population, the population is searched by using a gradient search optimization algorithm, and in the search, the motion direction and gradient change of the feature subset are calculated based on the classification performance feedback of the current feature subset on the random forest model, so that a reasonable and targeted local search direction can be formed near the pathological features and clinical performance features, the feature subset is quickly converged to the area with better prediction performance, and the optimal solution is screened by combining a local escape operator, so that the search efficiency is improved as a whole; then, dispersed foraging search is performed, the search process can simulate the dispersed and random foraging behavior of individuals in nature, the current solution is moderately randomly expanded in the feature space, the exploration range of the algorithm in the high-dimensional space composed of numerous clinical and pathological features is effectively expanded, the search process is prevented from prematurely falling into a local optimum, and through chaotic local search, the current optimal feature subset is finely adjusted by using the ergodicity and randomness of the chaotic sequence, so that the convergence speed is further accelerated on the basis of jumping out of the local optimum, so that the calculation efficiency is considered while the high prediction accuracy is ensured, and key features are identified for subsequent use. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A whole flowchart of a feature selection method for predicting inhaled allergy of children provided by the embodiment of the present application is shown in the figure. Figure 2 A selected frequency diagram of each feature in bDCGBO-RF of the feature selection method for predicting inhaled allergy of children provided by the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0017] In order to make the above objectives, characteristics and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application are described in detail below with reference to the drawings. In the following description, a large number of specific details are set forth in order to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the present application, so the present application is not limited by the specific embodiments disclosed below.

[0018] Allergic diseases in children, including atopic dermatitis, asthma, allergic rhinitis and food allergy, are increasingly prevalent chronic diseases worldwide, causing serious impact on children's health and quality of life. Inhalant allergy, caused by airborne allergens such as dust mites, pollen, mold and animal dander, is a common type of allergy and the main trigger of respiratory allergic diseases such as asthma and allergic rhinitis. In recent years, the incidence of inhalant allergy has been increasing year by year. For example, a study in Shanghai, China showed that the prevalence of childhood asthma was 10.2%, and the prevalence of rhinitis was as high as 54.1%, causing a heavy burden on the public health system and families. Inhalant allergy is an important stage in the evolution of allergic diseases in children, so early identification and intervention are of great significance.

[0019] With the rapid development of computer science and artificial intelligence technology, machine learning has been widely applied in the prediction of allergic diseases. Through the analysis of a large amount of clinical and environmental data, machine learning models can automatically identify potential risk factors and sensitization pathways, thereby achieving early prediction and individualized management of inhalant allergy in children. Compared with traditional statistical methods, machine learning has stronger data processing and pattern recognition capabilities, and can mine non-linear relationships in complex multi-dimensional data to provide more accurate decision support for clinical practice.

[0020] In the process of building a prediction model for inhalant allergy, due to the high dimensionality, feature redundancy and noise interference of clinical and environmental data, if all variables are directly used for modeling, it will often lead to increased model complexity and decreased prediction performance. Therefore, it is necessary to select key features that have a significant impact on the prediction results from the original data through feature selection methods, eliminate noise or irrelevant variables, and thus improve the prediction accuracy and stability of the model.

[0021] Feature selection can be generally divided into three categories: filter-based, wrapper-based and embedded-based methods. Filter-based methods evaluate the correlation between each feature and the target variable through statistical tests, regardless of the specific learning algorithm. Embedded methods implement feature selection by combining regularization techniques during model training. Wrapper-based methods directly use model performance as the evaluation standard, and evaluate the pros and cons of feature subsets by their actual prediction results. Compared with the other two methods, wrapper-based methods can more truly reflect the actual role of feature subsets in specific tasks, and therefore have obvious advantages in feature selection. However, since wrapper-based methods need to search among a large number of feature combinations, it is extremely difficult to exhaust all possible feature subsets in terms of computation, especially in the context of high-dimensional data. Therefore, it is of great significance to develop an efficient and robust feature selection method to improve the accuracy of predicting inhalant allergy in children.

[0022] In recent years, metaheuristic algorithms have been widely applied to feature selection tasks due to their excellent global search ability and adaptability. By simulating the optimization mechanisms of group cooperation or biological behavior in nature, metaheuristic algorithms can effectively balance global exploration and local exploitation in complex search spaces, thus finding a better feature subset. However, existing metaheuristic-based feature selection methods still face two major problems: first, the algorithm is prone to fall into local optimum in the later search stage, resulting in unstable feature selection results; second, the convergence speed of the algorithm is slow, making it difficult to balance computational efficiency while ensuring high prediction accuracy. To solve these problems, the present invention proposes an improved gradient optimizer, which is combined with random forest (RF) to build a wrapped feature selection model for predicting children's inhalation allergy, named bDCGBO-RF. In the present invention, a dispersed foraging strategy is introduced to expand the search range of the solution space and improve the global exploration ability of the algorithm, while a chaotic local search strategy is used to enhance the dynamic balance between global exploration and local exploitation, thereby improving the convergence performance of the optimization algorithm. Combined with the high accuracy and strong generalization ability of the random forest model in classification tasks, the present invention can more effectively identify key features and improve the accuracy of the inhalation allergy prediction model.

[0023] As shown in Figure 1 , the present invention provides a random forest optimizer and optimization algorithm combined children's inhalation allergy prediction method, specifically comprising: Step one: data set acquisition.

[0024] First, the original data set related to children's inhalation allergy is obtained; the data set contains multiple features, including pathological and clinical information, which are directly related to the risk of children's inhalation allergy.

[0025] Step two: feature filtering.

[0026] Four methods of mutual information, Relief, information gain and F-Score are used to preliminarily screen various clinical and environmental features (such as serum tlgE level, food sensitization, family allergy history, living environment, eating habits and growth and development status, etc.); these methods evaluate the importance of features from different angles, providing a high-quality candidate feature set for subsequent wrapped feature selection.

[0027] In this step, mutual information is used to evaluate the correlation strength between various clinical and environmental features (such as serum tlgE level, food sensitization, family allergy history, living environment, eating habits and growth and development status, etc.) and the presence of inhalation allergy, and its calculation formula is as follows: .

[0028] Among them: represents the feature with the class , respectively; the higher the MI value, the stronger the association between the feature and the class label.

[0029] Relief is a sample-instance-based feature selection method that evaluates the importance of various clinical and environmental features (e.g., serum tlgE levels, food sensitization, family history of allergy, living environment, dietary habits, and growth and development status) by measuring the difference between inhaled allergy-positive samples and inhaled allergy-negative samples; the basic idea is to randomly select a sample and calculate the distance between it and its nearest neighbor of the same class and its nearest neighbor of the opposite class; the weight of each feature is updated based on these distances, and the calculation formula is: .

[0030] where: represents the weight of feature A, and represent the difference between sample and its nearest neighbor of the same class and its nearest neighbor of the opposite class on feature A; the greater the weight, the more significant the contribution of the feature to the classification task.

[0031] Information gain is based on the concept of entropy and is used to measure the degree of reduction in uncertainty of data classification given a certain clinical or environmental feature; its calculation formula is as follows: .

[0032] where: represents the entropy of the class label Y , and represents the conditional entropy of Y given the feature X; the greater the information gain, the greater the contribution of the feature to distinguishing between inhaled allergy-positive and inhaled allergy-negative samples.

[0033] 9-Score is a statistical test method that evaluates the importance of various clinical and environmental features (e.g., serum tlgE levels, food sensitization, family history of allergy, living environment, dietary habits, and growth and development status) by measuring the difference between inhaled allergy-positive samples and inhaled allergy-negative samples; its calculation formula is as follows: .

[0034] where: is the number of samples of the kth class, is the mean of the kth class, is the overall mean, is the overall mean,​is the variance of the k-th type; the larger the F-Score value is, the stronger the discriminative ability of the feature in distinguishing whether there is inhalant allergy.

[0035] The above four methods generate ranking results of feature importance from different perspectives, thereby enabling multi-dimensional evaluation of the relevance and representativeness of the features; finally, the top key features are selected from each filtering method, and are combined to form a candidate feature set; wherein, the value is ( the total number of features); this strategy integrates the evaluation results of multiple filtering methods to obtain a more comprehensive and robust understanding of feature importance, providing reliable input for the subsequent package feature selection stage, thereby improving the robustness and reliability of the overall feature selection process.

[0036] Step three: data set division.

[0037] The data set is divided into a training set and a test set, wherein the training set is used to construct the model, and the test set is used to evaluate the generalization ability of the model.

[0038] Step four: initialization stage.

[0039] In this step, the population is initialized; the population size is set to , wherein each individual represents a coding vector of a candidate feature subset scheme; the dimension of the vector is the total number of features in the child inhalant allergy data set , which includes pathological indicators and clinical manifestations, all of which are related to the risk of child inhalant allergy; in the individual , each dimension corresponds to a feature in the data set with the same serial number, and the value is in the interval ; this value is used to represent the selection tendency and potential importance of the feature in the current scheme: a value equal to 1 indicates that the feature may have a high contribution to predicting the risk of child inhalant allergy, and the tendency is retained by the model; a value equal to 0 indicates that the feature has a low contribution in the current iteration, and the tendency is eliminated; to ensure the diversity of the search space, the initial values of all dimensions of all individuals are randomly generated in the interval , and the specific initialization process is as follows: .

[0040] wherein: represents the value of the k-th candidate feature subset in the i-th dimension, is the value of the i-th candidate feature subset in the k-th dimension, and is the value of the i-th candidate feature subset in the k-th dimension, and is the value of the i-th candidate feature subset in the k-th dimension, and a randomly generated value within the range of [0, 1]; next, in the subsequent search process, the present application determines whether each feature related to children's inhalant allergy is selected by a transformation function; the calculation method of the transformation function is as follows: .

[0041] .

[0042] wherein: represents the transformed position; represents the current candidate feature subset; is a random number within the range of [0, 1]; represents the transformation function.

[0043] Step five: gradient search optimization algorithm search.

[0044] After the population initialization, the gradient search optimization algorithm search is performed; the algorithm realizes the dynamic balance between global search and local development through the gradient search strategy and the local escape operator; first, the gradient search rule (GSR) is performed, which uses the motion direction (DM) information of individuals in the feature space to determine the influence trend of feature addition and deletion on the prediction accuracy of the children's inhalant allergy prediction model according to the performance feedback of the current feature subset; in this way, the algorithm can form a reasonable and targeted local search direction near the pathological and clinical manifestation features that are significantly related to children's inhalant allergy risk, so as to make the feature subset quickly converge to the area with better prediction performance; specifically, the potential position update formula of the current individual is as follows: .

[0045] .

[0046] .

[0047] .

[0048] wherein: and are random numbers within the range of [0, 1]; and respectively represent the value of the mth candidate feature subset and the current optimal feature subset in the mth dimension in the candidate feature subset set; and are new vectors generated based on GSR and DM; the parameter is used to balance the global exploration and local development stages of the algorithm.

[0049] The mathematical definition of GSR is as follows:​ .

[0050] .

[0051] .

[0052] .

[0053] wherein: represents a random number subject to a standard normal distribution; is a decimal number within the range of ; ; ; ; ; ; and represent the best solution and the worst solution in the population, respectively; is an integer randomly selected from the interval and is different from each other; meanwhile, ; ; .

[0054] The mathematical definition of the motion direction term DM is as follows: .

[0055] .

[0056] .

[0057] wherein: and have the same definition, represents a random number within the range of .

[0058] Next, a local escape operator (LEO) is executed; the operator generates a new feature subset solution by combining the current global optimal feature subset , two candidate solutions with good prediction performance , , and two random solutions , ; the mathematical expression thereof is as follows: .

[0059] .

[0060] .

[0061] .

[0062] where: is a uniform random number in the interval [0, 1], obeys a normal distribution with mean 0 and standard deviation 1; and is a random number in the interval [0, 1].

[0063] The parameters are defined as follows: .

[0064] .

[0065] .

[0066] where: is a binary parameter taking values 0 or 1; when , ; otherwise .

[0067] is a new solution generated within the domain of the feature space, defined as follows: .

[0068] .

[0069] where: is defined as ; is a randomly selected subset of solutions from the population; and are the lower and upper bounds of the search space, respectively, denotes the dimensionality of the feature space, i.e., the total number of pathological and clinical features contained in the dataset of children's inhalant allergy.

[0070] Step six: scatter-hunt search.

[0071] After the preliminary optimization results are obtained by the gradient search optimization algorithm, a dispersed foraging search process is performed to further improve the quality of the feature subset. This process simulates the dispersed and random foraging behavior of individuals in nature, moderately expands the current solution in the feature space, enhances the search range, and increases the diversity of candidate feature combinations. In this feature selection task, the dispersed foraging search can help the algorithm break free from the limitation of refining around a local superior feature set, allowing it to explore more pathological indicators and clinical symptom parameters that may have potential interactions with the risk of children's inhalation allergy. By introducing structural random disturbance and controllable dispersed movement, the dispersed foraging search helps to improve the diversity among individuals in the population, avoid premature convergence to a suboptimal feature set, and improve the comprehensive performance of the final selected feature subset in terms of prediction performance and biological interpretability. The mathematical model of this strategy is defined as follows:

[0072] .

[0073] .

[0074] .

[0075] wherein: and represent the positions of the th candidate feature subset in the current iteration and the next iteration, respectively; and represent two different individuals randomly selected from the population, and satisfy ; the parameter is a step control parameter, which follows a normal distribution with a mean of 0.5 and a standard deviation of 0.1; the adaptive diffusion rate (DR) is used to control the randomness of the search, with a value range of , represents the current number of function evaluations, represents the maximum number of function evaluations; by dynamically adjusting , the algorithm can maintain strong exploration ability in the early search stage and gradually enhance the exploitation ability in the later stage, thereby achieving an adaptive balance between exploration and exploitation.

[0076] Step seven: chaotic local search.

[0077] Subsequently, the chaotic local search is performed again; the chaotic local search can fine-tune the pathological features and clinical indicators that have been preliminarily screened out and are highly related to the risk of inhalation allergy in children, thereby further improving the ability of the model to identify key sensitization factors; at the same time, the introduction of the chaotic sequence can effectively avoid the search process from falling into a fixed mode or premature convergence, so that the algorithm can continuously mine potential high-contribution feature combinations in the local area and improve the prediction performance of the final feature selection results. The mathematical model of the chaotic mapping adopts the form of Logistic mapping, which is defined as follows:

[0078] .

[0079] wherein: , and ; when the control parameter , the Logistic function enters a fully chaotic state, thereby being able to generate a uniformly distributed chaotic sequence.

[0080] Based on the mapping, the chaotic local search function can be expressed as: .

[0081] wherein: and are the lower and upper bounds of the search space, respectively; the chaotic vector is mapped from the chaotic variable ; using the chaotic vector and the current optimal solution , the formula for generating a new candidate feature subset is as follows: .

[0082] .

[0083] wherein: represents the new candidate feature subset; by introducing the chaotic mechanism, the algorithm can continue to maintain the diversity of solutions and the ability to jump out of local optimality in the later search stage, thereby further improving the overall optimization performance.

[0084] Step eight: fitness evaluation and update.

[0085] The fitness evaluation and update process is a crucial step in the feature selection model of this invention. Its role is to determine whether each candidate feature subset solution has the value of being retained to the next iteration, or should be replaced by a better-performing solution. Fitness evaluation quantifies the classification performance of each feature subset combined with the childhood inhalation allergy prediction model to measure the effectiveness of the feature combination in the prediction task. Specifically, this invention aims to maintain or improve the model's prediction accuracy while reducing the number of features. During the evaluation process, not only the overall classification accuracy of the feature subset for predicting childhood inhalation allergy risk is considered, but also... The algorithm also focuses on the stability of the model in recognizing key clinical symptoms and pathological indicators. When a subset of features can maintain high predictive performance with fewer features, the solution is considered to have higher fitness and is preferentially retained or used as the update benchmark. Conversely, if a solution provides insufficient predictive performance or contains a large number of redundant features, its fitness is low and it will be replaced by a better solution during the update process. Through this fitness evaluation and update mechanism, the algorithm can gradually converge to an optimal subset of features that has both strong predictive ability and reflects the key sensitizing factors of inhaled allergies in children, thereby improving the clinical interpretability and practical application value of the model.

[0086] This invention employs a fitness function that comprehensively considers classification accuracy and the number of features to balance accuracy and model complexity; it uses Randomized Randomized Forecasting (RF) to calculate the prediction error for childhood inhalation allergies, while also considering the number of features; the fitness function is defined as: .

[0087] in: Indicates classification accuracy; The number of feature subsets; For feature dimensions; and The weighting parameters are set to 0.95 and 0.05 respectively.

[0088] Step 9: After the loop ends, output the population. .

[0089] The loop termination and output phase is the final stage of the entire optimization process, marking the end of the algorithm's search. By updating the position and fitness of individuals, the algorithm continuously improves the quality of the candidate feature subset, ultimately finding the optimal feature subset; when the number of iterations reaches a preset maximum value (…), the algorithm continues to improve the quality of the candidate feature subset. When the algorithm fails to reach the optimal feature set, it will terminate and output the current optimal feature subset. In the task of predicting inhaled allergies in children, the optimal solution is the optimal feature subset selected after algorithm optimization. This feature subset is obtained by minimizing the number of features and maximizing the accuracy of predicting inhaled allergies in children, thereby ensuring the efficiency and accuracy of the model.

[0090] The final output includes the optimal feature subset and the classification model trained based on the subset; by using RF to predict the test set, the corresponding classification results can be obtained, and the model performance can be evaluated; in order to comprehensively evaluate the model performance, the algorithm outputs multiple performance indicators such as accuracy, sensitivity, F value and area under ROC curve, etc., as shown in Tables 1, 2 and Figure 2 The experiments shown in Tables 1, 2 and 3 comprehensively reflect the performance of the model in predicting children's inhalation allergy, and effectively verify the effectiveness of the selected feature subset and the prediction ability of the model.

[0091] Table 1 Comparison between bDCGBO-RF and each meta-heuristic algorithm-based feature selection prediction method Table 2 Comparison between bDCGBO-RF and each traditional prediction method The above-described embodiments only express several embodiments of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of protection of the present application. Therefore, the scope of protection of the patent of the present application should be subject to the appended claims.

Claims

1. A feature selection method for predicting inhaled allergies in children, characterized in that, Includes the following steps: Obtain a feature dataset of pediatric patients associated with inhalation allergies, the feature dataset including pathological features and clinical features; The population is initialized based on the feature dataset. The gradient search optimization algorithm is used to determine the local search direction of the pathological and clinical features associated with childhood allergies. Based on the local search direction, individuals in the initialized population are guided to perform a global search in the feature space to obtain the globally optimal feature subset, two candidate solutions, and two random solutions. The local escape operator is then used to combine the globally optimal feature subset, the two candidate solutions, and the two random solutions to obtain multiple new feature subsets. Using a dispersed foraging search algorithm, the expansion direction of the interaction between pathological and clinical features associated with childhood allergies is determined, and multiple new feature subsets are guided to expand randomly in the feature space according to the expansion direction to obtain multiple interactive feature subsets. Using a chaotic local search algorithm, multiple interactive feature subsets are guided to perform local perturbation search in the feature space to obtain multiple final feature subsets. The fitness of multiple final feature subsets is evaluated, and the search is repeated until the maximum number of iterations is reached to obtain the optimal feature subset.

2. The feature selection method for predicting inhaled allergies in children according to claim 1, characterized in that, The search in the gradient search optimization algorithm includes: The population is represented as The population size is Each individual This represents the encoding vector of a candidate feature subset scheme, where the dimension of the vector is the total number of features contained in the candidate feature dataset. During the gradient search optimization algorithm search, a gradient search rule (GSR) search is performed. GSR utilizes the individual's motion direction (DM) information in the feature space and, based on the performance feedback of the current feature subset on the inhalation allergy prediction model, determines the trend of the impact of feature additions or deletions on prediction accuracy; then, the current individual... The potential location update formula is expressed as: ； ； ； ； in: and A random number within the interval [0,1]; and They represent the first and second elements in the candidate feature subset set, respectively. The values ​​of candidate feature subsets and the current optimal feature subset in the m-th dimension; and A new vector generated based on GSR and DM; parameters Used for the global exploration and local development phases of the balancing algorithm; The mathematical definition of GSR is expressed as: ； ； ； ； in: Represents a random number that follows a standard normal distribution; for Decimals within a range; ; ; ; ; ; and Let these represent the optimal and worst solutions in the population, respectively. From Randomly selected distinct integers within the interval; ; ; ; The mathematical definition of the direction of motion term DM is expressed as: ； ； ； in: and Same definition express A random number within a given range.

3. The feature selection method for predicting inhaled allergies in children according to claim 2, characterized in that, The search process for the local escape operator includes: After the gradient search optimization algorithm performs the search, the local escape operator combines the current globally optimal feature subset. Two candidate solutions with good prediction performance , and two random solutions , Generate new feature subset solutions Its mathematical expression is: ； ； ； ； in: for Uniformly random numbers within the interval It follows a normal distribution with a mean of 0 and a standard deviation of 1; and for Random numbers within a certain range; parameter The definition is expressed as: ； ； ； in: A binary parameter that takes the value 0 or 1; when hour, ;otherwise ; It is a new solution generated within the domain of the feature space, represented as: ； ； in: Definition and same; The solution is a randomly selected subset of features from the population. and These are the lower and upper bounds of the search space, respectively; This represents the dimension of the feature space.

4. The feature selection method for predicting inhaled allergies in children according to claim 3, characterized in that, The search process of the dispersed foraging search includes: Dispersed foraging search simulates the dispersed and random foraging behavior of individuals in nature, and randomly expands the current solution in the feature space to enhance the search range and increase the combination of candidate features; The dispersed foraging search is represented as: ； ； ； in: and They represent the first The positions of each candidate feature subset in the current iteration and the next iteration; and Let represent two distinct individuals randomly selected from the population that satisfy . ;parameter The step size control parameter follows a normal distribution with a mean of 0.5 and a standard deviation of 0.1; adaptive diffusivity. DR Used to control the randomness of the search, with a value range of [value range missing]. ; Indicates the number of times the function has been evaluated; This indicates the maximum number of function evaluations.

5. The feature selection method for predicting inhaled allergies in children according to claim 4, characterized in that, The search process of the chaotic local search includes: Chaotic local search performs local perturbation and fine-tuning on the pathological features and clinical indicators related to inhalation allergy that have been initially screened; Chaotic local search utilizes the Logistic chaotic map to generate chaotic sequences, which is represented as: ； in: ,and When control parameters When this happens, the Logistic function enters a completely chaotic state, thus generating a uniformly distributed chaotic sequence; Based on the Logistic chaotic mapping, the chaotic local search function is expressed as: ； in: and These represent the lower and upper bounds of the search space, respectively; the chaotic vector. From chaotic variables Obtained by mapping; Using chaotic vectors Compared with the current optimal solution The formula for generating a new candidate feature subset is expressed as: ； ； in: This represents a new subset of candidate features.

6. The feature selection method for predicting inhaled allergies in children according to claim 1, characterized in that, After the feature dataset is obtained, the method further includes feature filtering of the feature dataset, including: Feature filtering of the feature dataset involves filtering pathological and clinical features using mutual information, relief, information gain, and F-score. The mutual information is used to assess the strength of the association between various clinical and environmental characteristics and the presence of inhalation allergies; The Relief measure assesses the importance of various clinical and environmental characteristics by measuring the difference between samples with and without inhaled allergies. The information gain is based on the concept of entropy and is used to measure the degree to which the uncertainty of data classification is reduced given a certain clinical or environmental characteristic. The F-Score is used to assess the importance of various clinical and environmental characteristics by measuring the difference between samples with and without inhaled allergies.

7. A feature selection device for predicting inhaled allergies in children, characterized in that, include: The data acquisition module is used to acquire a feature dataset of pediatric patients associated with inhalation allergies, the feature dataset including pathological features and clinical features; The feature selection module initializes the population based on the feature dataset, uses a gradient search optimization algorithm to determine the local search direction for pathological and clinical features associated with childhood allergies, and guides individuals in the initialized population to perform a global search in the feature space based on the local search direction to obtain the globally optimal feature subset, two candidate solutions, and two random solutions; and uses a local escape operator to combine the globally optimal feature subset, two candidate solutions, and two random solutions to obtain multiple new feature subsets. Using a dispersed foraging search algorithm, the expansion direction of the interaction between pathological and clinical features associated with childhood allergies is determined, and multiple new feature subsets are guided to expand randomly in the feature space according to the expansion direction to obtain multiple interactive feature subsets. Using a chaotic local search algorithm, multiple interactive feature subsets are guided to perform local perturbation search in the feature space to obtain multiple final feature subsets. The fitness of multiple final feature subsets is evaluated, and the search is repeated until the maximum number of iterations is reached to obtain the optimal feature subset.

8. An electronic device, characterized in that, include: Memory and processor; The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the steps of the feature selection method for predicting inhalation allergies in children as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, Used to store a computer program, which, when executed by a processor, implements the steps of a feature selection method for predicting inhalation allergies in children as described in any one of claims 1 to 6.

Images