High-dimensional medical feature selection method, system and device
By combining moss growth optimization algorithms, spore diffusion strategies, and differential evolution-guided feature selection methods, the efficiency and accuracy issues of feature selection in high-dimensional medical data are solved, achieving efficient feature subset selection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BIG DATA & INFORMATION TECH RES INST OF WENZHOU UNIV
- Filing Date
- 2026-01-13
- Publication Date
- 2026-04-24
AI Technical Summary
Existing metaheuristic algorithms struggle to efficiently and accurately select features in high-dimensional medical data, exhibiting problems such as convergence speed decreasing exponentially with dimensionality, traditional mutation operators being prone to ineffective searches and getting trapped in local minima, and premature convergence.
We employ a moss growth optimization algorithm combined with a spore dispersal strategy, a dual reproduction search, and a differential evolution guidance strategy. Feature selection is performed based on a comprehensive score of individual fitness value and Euclidean distance, dynamically balancing global exploration and local development to avoid excessively high similarity of feature subsets.
It improves the accuracy and efficiency of high-dimensional medical feature selection, overcomes the curse of dimensionality, and provides an efficient feature selection scheme.
Smart Images
Figure CN121919554A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical data processing technology, specifically to a method, system, and device for selecting high-dimensional medical features. Background Technology
[0002] In recent years, the application of machine learning methods in the medical field has shown a significant trend of widespread adoption. Compared with traditional analysis methods, machine learning algorithms have unique advantages in capturing subtle patterns and potential correlations in complex datasets. Feature selection, as a core step in machine learning and data mining, aims to filter out highly relevant subsets of features to improve classifier performance. Medical datasets are often high-dimensional, which poses a significant challenge to dimensionality reduction techniques due to their high computational complexity. Such datasets often contain a large number of irrelevant, redundant, and noisy features, making it difficult for traditional machine learning methods to effectively handle such a large feature set. In these scenarios, model performance often falls short of expectations, highlighting the increasing importance of feature selection as a preprocessing step: by systematically removing irrelevant, redundant, or noisy features, this technique can effectively optimize evaluation metrics and extract a more representative subset of features from the original data, laying the foundation for subsequent modeling and analysis.
[0003] Currently, based on their implementation mechanisms, feature selection techniques can be systematically divided into three typical paradigms: filtering, wrapping, and embedded. Filtering methods construct feature evaluation index systems independent of specific learning algorithms, such as information gain or mutual information based on information theory, or variance thresholds based on statistics. Features are quantified, ranked, and screened according to general evaluation criteria to assess feature importance. Although filtering methods can analyze features independently without involving machine learning models, there is a deviation between the optimization objective of the selected feature subset and the actual performance of the classification model. Embedded methods deeply integrate the feature selection process into the classifier training framework, by adding features to the model's objective function. Structural constraints, such as L1 regularization, decision tree pruning strategies, or adaptive weighting mechanisms, can be introduced to achieve implicit selection and optimization of feature subsets. Although embedded methods can reveal complex relationships between features, they require consideration of complex parameters and structures and are susceptible to the influence of the machine learning model itself. Wrapping methods have attracted much attention due to their simplicity and ease of operation. However, when existing traditional metaheuristic optimization algorithms such as genetic algorithms and particle swarm optimization are applied to feature selection problems, their search strategies fail to fully consider the inherent structural characteristics of the feature space, making them susceptible to the "curse of dimensionality" in high-dimensional data environments—the algorithm's performance decreases exponentially as the feature dimension increases.
[0004] In summary, while metaheuristic algorithms can quickly find optimal or near-optimal solutions in the development of wrap-around feature selection, existing metaheuristic algorithms face three major challenges when solving feature selection problems: convergence speed decreases exponentially with dimensionality, traditional mutation operators are prone to ineffective search and getting trapped in local minima, and premature convergence is common. These challenges make it difficult to efficiently and accurately select high-dimensional medical features. Summary of the Invention
[0005] To address the shortcomings of existing search strategies that fail to fully consider the inherent structural characteristics of the feature space, making them susceptible to the "curse of dimensionality" in high-dimensional data environments and hindering the efficient and accurate selection of high-dimensional medical features, this invention proposes a method, system, and apparatus for selecting high-dimensional medical features, thereby solving the problems existing in the prior art.
[0006] A method for selecting high-dimensional medical features includes the following steps: Obtain a medical dataset and initialize a population for searching the optimal feature subset in the medical dataset; the population consists of multiple individuals, where each individual is a feature vector with a length equal to the dimension of the dataset; Each individual in the current population is binary-encoded using a conversion function; The population after binary encoding is searched using a moss growth optimization algorithm; a comprehensive score is obtained by calculating the fitness value of each individual in the searched population and its Euclidean distance from the current optimal solution; and crossover iteration is performed on individuals whose comprehensive scores exceed a preset threshold to update the population. A differential evolution-guided search strategy is applied to the updated population to generate a trial vector for each individual. The generated trial vector is compared with the original individual, and a greedy selection strategy is used to retain the best individual to generate the next generation population. After multiple iterations, the maximum number of iterations is reached, and the final best individual is output. The selected medical feature combination is based on the feature subset represented by the binary encoding vector corresponding to the final optimal individual.
[0007] Furthermore, the binary encoding of each individual in the current population using a conversion function is specifically represented as follows: ; ; in, Indicates the first t The first iteration generated i The first individual j 1 eigenvalue, Indicates the first t The generation in the +1st iteration i The first individual jA binary encoded value, T This represents the conversion function.
[0008] Furthermore, the moss growth optimization algorithm includes a spore dispersal strategy and a dual reproduction search strategy; The update rule for the spore dispersal strategy is specifically as follows: ; in, A random number that is neither greater than 1 nor less than 0. Choose a threshold for the step size. Indicates the first i An updated population, Indicates the first i One original population, Indicates the steady wind step length. Corresponding to the turbulent wind step length, As the direction of population evolution; The update rules for the dual-reproduction search strategy are as follows: ; in, A random number that is neither greater than 1 nor less than 0. Choose a threshold for the step size. Indicates the first i An updated population, Indicates the first i One original population, Indicates the steady wind step length. Corresponding to the turbulent wind step length, This indicates the direction of population evolution.
[0009] Furthermore, the fitness value of each individual in the population is expressed as: ; ; ; in, This indicates the classification accuracy of the KNN classifier. For classification error rate Indicates the first Fitness values of each population Indicates the number of selected features. This represents the total number of features in the dataset. and It is a weighting coefficient used to balance the contribution of two different factors to fitness.
[0010] Furthermore, the comprehensive score is obtained by calculating the fitness value of each individual in the searched population and its Euclidean distance to the current optimal solution, specifically including the following steps: Calculate the Euclidean distance between each individual in the population after the search and the current optimal solution: ; in Indicates the current individual, This indicates the location of the optimal solution recorded. Indicates population size; Normalize the Euclidean distance and fitness values: ; ; in, For the first The distance values of an individual in the original distance set. This represents the fitness value of an individual. and These represent the normalized distance and fitness values, respectively; It is a positive number; Calculate the individual's overall score based on the normalized Euclidean distance and fitness value: ; in, These are the weighting coefficients.
[0011] Furthermore, the process of performing crossover iterations on individuals whose comprehensive scores exceed a preset threshold to update the population is specifically represented as follows: ; in This represents the solution vector of the individual with the highest score. The solution vector representing the individual with the lowest score; It is a random binary vector; These are random values generated from a standard normal distribution. Let be the individual solution vector randomly selected from the population, and be . , They jointly participate in calculations to generate a new population. .
[0012] Furthermore, the differential evolution-guided search strategy applied to the updated population generates a trial vector for each individual, specifically as follows: ; in This represents the population currently undergoing iteration. This is the scaling factor. This represents the optimal solution obtained through the cryptic mechanism in the moss growth optimization algorithm. Obtained by averaging from the population, the values range from 1 to the total number of features. N between, , They represent the first and the Population.
[0013] The present invention also includes a high-dimensional medical feature selection system, comprising: The acquisition module is used to acquire a medical dataset and initialize a population for searching the optimal feature subset in the medical dataset; the population consists of multiple individuals, where each individual is a feature vector with a length equal to the dimension of the dataset; The encoding module is used to encode each individual in the current population into binary form using a conversion function. The update module is used to search the binary-encoded population using a moss growth optimization algorithm; by calculating the fitness value of each individual in the searched population and the Euclidean distance from the current optimal solution, a comprehensive score is obtained; and crossover iteration is performed on individuals whose comprehensive scores exceed a preset threshold to update the population. The generation module is used to implement a differential evolution-guided search strategy on the updated population, generating a trial vector for each individual; the generated trial vector is compared with the original individual, and a greedy selection strategy is used to retain the best individual to generate the next generation population; after multiple iterations, until the maximum number of iterations is reached, the final best individual is output; The selection module is used to select a subset of medical features based on the feature subset represented by the binary encoding vector corresponding to the final optimal individual.
[0014] The present invention also includes a high-dimensional medical feature selection computer device, comprising: a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the high-dimensional medical feature selection method.
[0015] The present invention also includes a readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, are used to perform the steps of the high-dimensional medical feature selection method.
[0016] This invention provides a high-dimensional medical feature selection method, which has the following beneficial effects: This invention integrates a fitness and distance scoring strategy with a differential evolution guidance strategy into the original moss growth optimization algorithm. Position updates are based on the individual's fitness value and the distance to the optimal solution. The fitness and distance strategy improves population diversity and avoids excessive similarity of feature subsets by balancing individual fitness with the distance to the optimal solution. The differential evolution guidance strategy improves convergence efficiency by guiding the global search direction through elite individuals. This method overcomes the high-dimensional efficiency bottleneck of traditional metaheuristic algorithms. By considering the inherent structural characteristics of the feature space, it dynamically balances global exploration and local development to overcome premature convergence and avoids the susceptibility to the "curse of dimensionality" in high-dimensional data environments, providing an efficient and accurate solution for medical feature selection. Attached Figure Description
[0017] Figure 1 This is a flowchart of the high-dimensional medical feature selection method in an embodiment of the present invention; Figure 2 This is a schematic diagram comparing the fitness value, accuracy, time, and number of selected features of BFPMGO with various algorithms in this embodiment of the invention: Figure 2 Figures a, b, c, d, e, and f in the figure represent the comparison diagrams between BFPMGO and various algorithms in terms of fitness value, accuracy, time, and number of selected features on the blood, SBRCT, Leukemia, Prostate_Tumor, IonosphereEW, and Colon datasets, respectively. Figure 3 This is a graph showing the fitness decrease of BFPMGO and various algorithms during the feature selection iteration process in the embodiments of the present invention. Figure 3 Figures a, b, c, and d in the figure represent the fitness decrease curves of BFPMGO and various algorithms during the feature selection iteration process on the Prostate_Tumo, Colon, DLBCL, and Leukemia2 datasets, respectively. Detailed Implementation
[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.
[0019] This invention proposes a high-dimensional medical feature selection method, such as... Figure 1 As shown, the specific steps include: S1. Obtain the medical dataset and divide it into a training set and a validation set.
[0020] Table 1 Dataset Information S2. Calculate the fitness value of the current population: Based on the training sample set obtained from feature extraction, let the initial training sample population be set. =( , ..., ), i =1,2,..., N , j =1,2,..., D , t =0, where, N The number of individuals in the training sample. D The number of dimensions for each training sample. Indicates the first t The population obtained in each iteration, the set of feature vectors in the medical dataset, corresponds to the "sample set in the population." Each feature vector describes the characteristics of a single individual in the population. Indicates in t In the nth iteration i The individual's first j 1 eigenvalue, t The current iteration number is represented; the individual represents a binary feature vector with a length equal to the dimension of the dataset. Each element (0 or 1) in the vector corresponds to a feature in the original dataset and is used to indicate whether the feature is selected. If the element is 1, it means that the corresponding feature is selected and will participate in subsequent model training; if the element is 0, it means that the corresponding feature is removed and will not participate in subsequent model training.
[0021] Then, relevant features are selected and fitness values are calculated using a transformation function; the fitness values of the population are calculated using binary representation, specifically using the following formula: ; ; ; Among them, parameters Set it to 0.95. This indicates the classification accuracy of the KNN classifier. This represents the fitness value of an individual. For classification error rate, Indicates the number of selected features. This represents the total number of features in the dataset. and It is a weighting coefficient used to balance the contribution of two different factors to fitness.
[0022] The conversion function is as follows: ; ; in, Indicates the first t The first iteration generated i The first individual j 1 eigenvalue, Indicates the first t The generation in the +1st iteration i The first individual j A binary encoded value, T Both represent transformation functions.
[0023] S3. The population is searched using a moss growth optimization algorithm, with the spore diffusion search and update rules specifically as follows: ; in, A random number that is neither greater than 1 nor less than 0. Set the step size to 0.2 for the threshold selection. This indicates the update after using the spore diffusion search update rules. i A population of individuals, Indicates the first i One original population, Indicates the steady wind step length. Corresponding to the turbulent wind step length, The evolutionary winds of the population correspond to the following formulas: ; ; ; in for Total number of individuals for Internal correspondence The set of distances, This represents wind strength, which decreases linearly with each iteration. The specific expression is as follows: ; ; ; in One-quarter of the total number of dimensions For the first Random dimension index, The number of times the current function is evaluated. This represents the maximum number of evaluations.
[0024] S4. After the spore diffusion search in step S3, if specific conditions are met, the dual reproduction search strategy in the moss growth algorithm is used to generate locally optimized new individuals by simulating the sexual and vegetative reproduction strategies of moss. If the conditions are met after the spore diffusion search... Triggered afterward, among which This represents a random number between 0 and 1. If the condition is not met, the result of step S3 is retained and the double-reproduction search strategy is not performed. Its specific expression is as follows: ; ; ; in This indicates the update of the first generation through a dual-reproduction search strategy. i The first in the population j New individuals in dimensions Indicates the first i The first in the population j The original individual in the dimension, Indicates the wind direction. j The specific wind direction of the dimension, Indicates the propagation step size. This is a value set to 1 with a probability of 0.5 and otherwise to 0. Simultaneously, the occultation mechanism of the moss growth algorithm is used to record each... T The optimal solution for each generation is used to initialize individuals upon stopping, and its specific expression is as follows:
[0025] ; in T Setting it to 50 means the optimal solution is updated every fifty iterations. If the storage capacity is exceeded, the optimal solution will be replaced according to the first-in-first-out principle. For the first k · T The optimal individual of the generation.
[0026] S5. For the population after applying the moss optimization algorithm, perform a crossover operation based on the distance and fitness scores. The crossover is performed on the population based on fitness and the distance between each individual and the current optimal solution, where the distance is Euclidean distance, specifically: ; in Indicates the current individual, This indicates the location of the optimal solution recorded. The population size is represented by the formula below, which is then used to normalize the distance and fitness values. ; ; in, For the first The distance values of an individual in the original distance set. For the first i The fitness value of each individual; and These represent the normalized distance and fitness values, respectively; The value is a small positive number to avoid division by zero errors. Then, additional individual operations are performed by calculating the individual's overall score, using the following formula:
[0027] ; in, The weighting coefficient (set to 0.5) is used to balance the normalized fitness. With normalized distance The importance of this. When the score exceeds the mean by more than or equal to two standard deviations, a crossover operation is performed to update the population, as described in the following formula:
[0028] ; in This represents the solution vector of the individual with the highest score. The solution vector representing the individual with the lowest score; It is a random binary vector; These are random values generated from a standard normal distribution. Let be the individual solution vector randomly selected from the population, and be . , They work together to calculate and generate new solutions.
[0029] Then, a further step is taken to guide the population through differential evolution based on the cryptic mechanism of the moss growth algorithm. The update formula is as follows: ; in This represents the population currently undergoing iteration. The scaling factor has a value between 0 and 1. The optimal solution is obtained through the hidden generation mechanism. Obtained by averaging from the population, the values range from 1 to... N between, , They represent the first and the Population.
[0030] S6. Fitness value evaluation and updating of individual historical records. To reasonably evaluate the performance of each algorithm, the following metrics were used: mean fitness value (MF), classification error rate (MER), mean number of selected features (MFS), and mean running time (ART).
[0031] Smaller MF and MER values, along with a lower MFS value, indicate better algorithm performance. MER is the average error rate across all individuals in the test data, and ART is the average time taken by each algorithm after 10 independent runs. The binarized distance-fitness-optimized moss growth algorithm (BFPMGO) was compared with other algorithms using a Wilcoxon rank-sum (WRS) test, with a significance level of 0.05. Furthermore, a '+' indicates that the compared algorithm is significantly better than BFPMGO, a '-' indicates that the compared algorithm is significantly worse than BFPMGO, and an '=' indicates that there is no significant difference between the compared algorithm and BFPMGO.
[0032] S7. If the maximum number of iterations is reached, the loop ends, the population optimal solution is output, and the final combination of medical features is obtained based on the population optimal solution.
[0033] The method of this invention is compared with other optimization algorithms, including the binarized gray wolf optimization algorithm (BGWO), the binarized moth-to-fire optimization algorithm (BMFO), the binarized particle swarm optimization algorithm (BPSO), the binarized antlion hunting optimization algorithm (BALO), the binarized bat optimization algorithm (BBA), the binarized sea squirt optimization algorithm (BSSA), the binarized whale optimization algorithm (BWOA), and the binarized slime mold predation algorithm (BSMA). Figure 2 The graphs a, b, c, d, e, and f in the figure correspond to the ranking of the proposed method in the four dimensions of feature number, fitness, accuracy, and time consumption. Figure 3 Figures a, b, c, and d in the figure correspond to the convergence curves of the fitness of the method proposed in this invention as the number of iterations increases. Figure 2 The blood, SBRCT, Leukemia, Prostate Tumor, IonosphereEW, Colon and Figure 3 The Prostate_Tumo, Colon, DLBCL, and Leukemia2 in the table correspond to the datasets used in Table 1. Figure 2 , Figure 3The thicker lines represent the method proposed in this invention, while the remaining dashed lines represent similar comparative algorithms, including the binarized gray wolf optimization algorithm (BGWO), the binarized moth-to-fire optimization algorithm (BMFO), the binarized particle swarm optimization algorithm (BPSO), the binarized antlion hunting optimization algorithm (BALO), the binarized bat optimization algorithm (BBA), the binarized sea squirt optimization algorithm (BSSA), the binarized whale optimization algorithm (BWOA), and the binarized slime mold predation algorithm (BSMA).
[0034] This invention proposes a medical feature selection method that integrates a fitness and distance scoring strategy with a differential guidance strategy on top of the original moss growth optimization algorithm. Position updates are performed based on the individual's fitness value and the distance to the optimal solution. The fitness and distance strategy improves population diversity and avoids excessive similarity among feature subsets by balancing individual fitness with the distance to the optimal solution. The differential guidance strategy guides the global search direction through elite individuals, improving convergence efficiency. This method overcomes the high-dimensional efficiency bottleneck of traditional metaheuristic algorithms by dynamically balancing global exploration and local exploitation to overcome premature convergence, providing an efficient and accurate solution for medical feature selection.
[0035] Based on the same inventive concept, this invention also proposes a high-dimensional medical feature selection system, comprising: The acquisition module is used to acquire the medical dataset and initialize a population for searching the optimal feature subset in the medical dataset. The population consists of multiple individuals, where each individual is a feature vector with a length equal to the dimension of the dataset.
[0036] The encoding module is used to encode each individual in the current population into binary form using a conversion function.
[0037] The update module is used to search the binary-encoded population using a moss growth optimization algorithm; by calculating the fitness value of each individual in the searched population and its Euclidean distance from the current optimal solution, a comprehensive score is obtained; individuals with comprehensive scores exceeding a preset threshold are subjected to cross-iteration operations to update the population.
[0038] The generation module is used to guide the search strategy through differential evolution on the updated population, generating a trial vector for each individual; the generated trial vector is compared with the original individual, and a greedy selection strategy is used to retain the best individual to generate the next generation population; after multiple iterations, until the maximum number of iterations is reached, the final best individual is output.
[0039] The selection module is used to select a subset of medical features based on the feature subset represented by the binary encoding vector corresponding to the final optimal individual.
[0040] The present invention also proposes a high-dimensional medical feature selection computer device, comprising: a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the high-dimensional medical feature selection method.
[0041] The present invention also proposes a readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, are used to perform the steps of a high-dimensional medical feature selection method.
[0042] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A method for selecting high-dimensional medical features, characterized in that, Includes the following steps: Obtain a medical dataset and initialize a population for searching the optimal feature subset in the medical dataset; the population consists of multiple individuals, where each individual is a feature vector with a length equal to the dimension of the dataset; Each individual in the current population is binary-encoded using a conversion function; The population after binary encoding is searched using a moss growth optimization algorithm; the comprehensive score is obtained by calculating the fitness value of each individual in the searched population and its Euclidean distance from the current optimal solution. Individuals whose overall scores exceed a preset threshold are subjected to crossover iterations to update the population; A differential evolution-guided search strategy is applied to the updated population to generate an experimental vector for each individual. The generated experimental vectors are compared with the original individuals, and a greedy selection strategy is used to retain the best individual to generate the next generation of the population. After multiple iterations and updates, until the maximum number of iterations is reached, the final optimal individual is output. The selected medical feature combination is based on the feature subset represented by the binary encoding vector corresponding to the final optimal individual.
2. The method for selecting high-dimensional medical features according to claim 1, characterized in that, The binary encoding of each individual in the current population using a conversion function is specifically represented as follows: ； ； in, Indicates the first t The first iteration generated i The first individual j 1 eigenvalue, Indicates the first t The generation in the +1st iteration i The first individual j A binary encoded value, T This represents the conversion function.
3. The method for selecting high-dimensional medical features according to claim 1, characterized in that, The moss growth optimization algorithm includes a spore dispersal strategy and a dual reproduction search strategy; The update rule for the spore dispersal strategy is specifically as follows: ； in, A random number that is neither greater than 1 nor less than 0. Choose a threshold for the step size. Indicates the first i An updated population, Indicates the first i One original population, Indicates the steady wind step length. Corresponding to the turbulent wind step length, As the direction of population evolution; The update rules for the dual-reproduction search strategy are as follows: ； in, A random number that is neither greater than 1 nor less than 0. Choose a threshold for the step size. Indicates the first i An updated population, Indicates the first i One original population, Indicates the steady wind step length. Corresponding to the turbulent wind step length, This indicates the direction of population evolution.
4. The method for selecting high-dimensional medical features according to claim 1, characterized in that, The fitness value of each individual in the population is represented as: ； ； ； in, This indicates the classification accuracy of the KNN classifier. For classification error rate, Indicates the first Fitness values of each population Indicates the number of selected features. This represents the total number of features in the dataset. and It is a weighting coefficient used to balance the contribution of two different factors to fitness.
5. The method for selecting high-dimensional medical features according to claim 1, characterized in that, The process of calculating the fitness value of each individual in the searched population and its Euclidean distance from the current optimal solution to obtain a comprehensive score includes the following steps: Calculate the Euclidean distance between each individual in the population after the search and the current optimal solution: ； in Indicates the current individual, This indicates the location of the optimal solution recorded. Indicates population size; Normalize the Euclidean distance and fitness values: ； ； in, For the first The distance values of an individual in the original distance set. This represents the fitness value of an individual. and These represent the normalized distance and fitness values, respectively; It is a positive number; Calculate the individual's overall score based on the normalized Euclidean distance and fitness value: ； in, These are the weighting coefficients.
6. The method for selecting high-dimensional medical features according to claim 1, characterized in that, The process of performing crossover iterations on individuals whose comprehensive scores exceed a preset threshold to update the population is specifically represented as follows: ； in This represents the solution vector of the individual with the highest score. The solution vector representing the individual with the lowest score; It is a random binary vector; These are random values generated from a standard normal distribution. Let be the individual solution vector randomly selected from the population, and be . , They jointly participate in calculations to generate a new population. .
7. The method for selecting high-dimensional medical features according to claim 1, characterized in that, The differential evolution-guided search strategy for the updated population generates a test vector for each individual, specifically as follows: ； in This represents the population currently undergoing iteration. This is the scaling factor. This represents the optimal solution obtained through the cryptic mechanism in the moss growth optimization algorithm. Obtained by averaging from the population, the values range from 1 to the total number of features. N between, , They represent the first and the Population.
8. A high-dimensional medical feature selection system, characterized in that, include: The acquisition module is used to acquire medical datasets and initialize a population for searching for the optimal feature subset in the medical datasets. The population consists of multiple individuals, each of which is a feature vector with a length equal to the dimension of the dataset. The encoding module is used to encode each individual in the current population into binary form using a conversion function. The update module is used to search the binary-encoded population using a moss growth optimization algorithm; a comprehensive score is obtained by calculating the fitness value of each individual in the searched population and its Euclidean distance from the current optimal solution. Individuals whose overall scores exceed a preset threshold are subjected to cross-iteration operations to update the population; The generation module is used to implement a differential evolution-guided search strategy on the updated population, generating an experimental vector for each individual; the generated experimental vector is compared with the original individual, and a greedy selection strategy is used to retain the best individual to generate the next generation population; After multiple iterations and updates, until the maximum number of iterations is reached, the final optimal individual is output. The selection module is used to select a subset of medical features based on the feature subset represented by the binary encoding vector corresponding to the final optimal individual.
9. A high-dimensional medical feature selection computer device, characterized in that, include: A memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the high-dimensional medical feature selection method according to any one of claims 1-7.
10. A readable storage medium, characterized in that, The readable storage medium stores a computer program, the computer program including program instructions, which, when executed by a processor, are used to perform the steps of the high-dimensional medical feature selection method according to any one of claims 1-7.