A feature selection method and system based on an improved slime optimization algorithm

Abstract

This invention discloses a feature selection method and system based on an improved slime mold optimization algorithm, belonging to the field of data feature selection technology. This invention uses K-means clustering to divide the population into several independent clusters and updates the individual positions based on the cluster center vectors, enabling the potential advantageous regions of the population to be more fully and continuously developed. An improved greedy selection method is used to obtain the instantaneous position update effectiveness and cumulative stagnation history of each individual, using individual information to guide the population's evolutionary direction. MSHDC is used to dynamically adjust the crossover probability of operators based on the population state, and by integrating the instantaneous position update effectiveness and cumulative stagnation history of each individual, a suitable strategy is selected to update the individuals, avoiding the algorithm from getting trapped in local optima and missing key features, and obtaining the optimal feature subset in the final feature selection result.

Description

Technical Field

[0001] This invention relates to the field of data feature selection technology, and in particular to a feature selection method, system, device and medium based on an improved slime mold optimization algorithm. Background Technology

[0002] Driven by the rapid development of digital infrastructure, high-dimensional data is becoming increasingly common, especially in the medical field, where it has shown significant scientific research value and clinical application potential. However, the high-dimensional nature of medical data also brings severe analytical challenges. When the number of features far exceeds the number of available observation samples, the curse of dimensionality not only leads to an exponential increase in computational complexity but also exacerbates the decline in model performance and the risk of overfitting. This has become a key bottleneck that must be overcome in high-dimensional data analysis.

[0003] To address this challenge, feature selection has emerged, with its core objective being to select the most informative subset of features while maintaining or even improving the prediction accuracy of the original model. Traditional feature selection methods, such as filtering, wrapping, and embedding methods, often perform poorly when dealing with high-dimensional medical data. Filtering methods are fast but struggle to detect interactions between features, resulting in lower accuracy; wrapping methods offer high accuracy but incur huge computational costs and are unsuitable for high-dimensional spaces; embedding methods rely on specific learning algorithms and have limited generalization capabilities. To address these challenges, feature selection methods based on metaheuristic algorithms have attracted widespread attention due to their powerful global search capabilities. Among them, the Slime Mould Algorithm (SMA), as a novel swarm intelligence optimization algorithm, has demonstrated some effectiveness in solving the feature selection problem.

[0004] The core update formula of the SMA algorithm relies on the relative position difference between the current optimal individual and randomly selected individuals to guide population evolution. This mechanism works well for low-dimensional, continuous optimization problems with relatively smooth solution space topography. However, feature selection in high-dimensional medical data is an ultra-high-dimensional discrete combinatorial optimization problem. The solution space topography is extremely rugged, filled with a massive number of local extrema of varying quality, and the feature subsets corresponding to these local extrema may be drastically different in clinical diagnostic significance. In this scenario, the SMA algorithm's reliance on the current optimal individual position causes the population to cluster towards a relatively good local solution, missing feature combinations that are initially not obvious but are actually extremely important, i.e., getting stuck in "local optima". At the same time, the SMA algorithm's mechanism of guiding population evolution by randomly selecting the relative position difference between individuals still cannot carry out refined local development in the later stages of iteration, making it difficult to effectively distinguish those subtle difference feature combinations that are decisive in clinical diagnosis. Ultimately, due to the imbalance between "exploration and development", key features are missed and clinical discriminative power is reduced, making it difficult to effectively select the most accurate key features from high-dimensional medical data. Summary of the Invention

[0005] This invention provides a feature selection method and system based on an improved slime mold optimization algorithm, which can solve the problems existing in the prior art.

[0006] This invention provides a feature selection method based on an improved slime mold optimization algorithm, comprising the following steps: Obtain the medical feature dataset to be processed; An initial solution set is generated based on a medical feature dataset, and an improved slime mold optimization algorithm is used to iteratively find the optimal feature subset: K-means clustering is used to divide the population in the current initial solution set into several independent clusters, and based on the cluster center vector of the cluster to which each individual in the initial solution set belongs and the difference between the cluster center vectors of other clusters, the individuals in the initial solution set are guided to perform directional mutation updates based on the current individual position when searching in the feature space, thereby obtaining multiple candidate solutions; For multiple candidate solutions, obtain the global population diversity state and global stagnation time of all individuals in the population, as well as the instantaneous position update validity and cumulative stagnation history of each individual; dynamically obtain a crossover probability based on the global population diversity state and global stagnation time, and combine the instantaneous position update validity and cumulative stagnation history of each individual to adaptively select a strategy from a variety of preset update strategies to update the individual positions in multiple candidate solutions, thus obtaining the final multiple candidate solutions; For the final multiple candidate solutions, obtain the fitness value of each solution, aim to minimize the fitness value of each solution, and repeat the search until the maximum number of iterations is reached to obtain the optimal feature subset corresponding to the optimal solution; Preferably, the directed mutation update includes: K-means clustering is used to divide the population in the current initial solution set into several independent clusters, and the individual position is mutated and updated based on the difference between the cluster center vector of the cluster to which each individual belongs in the initial solution set and the cluster center vector of other clusters. When guiding individual locations to mutate and update, a dynamic mutation probability is set based on the mutation guidance strategy CIM, which linearly decreases from a higher value to a lower value as the iteration progresses. When the number of random variables is greater than the preset first dynamic mutation probability, the difference vector between the cluster center of the individual's current cluster and the cluster center of another randomly selected cluster is used to guide cross-regional directional mutation of the individual. When the number of random variables is greater than the preset second dynamic mutation probability but less than the preset first dynamic mutation probability, a Gaussian random perturbation is superimposed on the cluster center of the individual, and exploratory updates are performed at the cluster center.

[0007] Preferably, the acquisition of the instantaneous position update validity and cumulative stagnation history of each individual includes: The greedy selection mechanism is used to obtain the instantaneous position update effectiveness and cumulative stagnation history of each individual in the population of multiple candidate solutions; When using the greedy selection mechanism, if the fitness value of a candidate solution is higher than the preset fitness value, the solution is accepted and used as the new position. At the same time, the instantaneous update validity of the individual is marked as successful, and the accumulated stagnation count is cleared to zero. If the fitness value of a candidate solution is lower than the preset fitness value, the solution is rejected and the original position is maintained. The instantaneous update validity is marked as failed, and the accumulated stagnation count is incremented by one.

[0008] Preferably, the acquisition of the final multiple candidate solutions includes: Based on the global population diversity state and global stagnation time, the crossover probability of the operator is obtained. The crossover probability of the operator is jointly determined by the diversity level represented by the average inter-individual distance of the population and the global stagnation time represented by the number of consecutive generations without updating the global optimal solution. Based on the crossover probability of the operator, and combined with the instantaneous position update effectiveness and cumulative stagnation history of the individual, an adaptive strategy is selected from a variety of preset update strategies to update the individual, thus obtaining multiple candidate solutions. The update strategy includes: for successful and non-stagnant individuals, an open-ended update is performed using the difference between the composite guiding vector and the current individual; for unsuccessful but non-stagnant individuals, a local exploration is performed using the difference between the random individual and the current individual; for successful but stagnant individuals, a guided jump exploration is performed using the difference between the random individual and the composite guiding vector; and for unsuccessful and stagnant individuals, a strongly random perturbation exploration is performed using the difference between two random individuals.

[0009] Preferably, obtaining the fitness value of each solution includes: Each individual vector within the final multiple candidate solutions is converted into a binary vector using a transformation function. The KNN model is then used to calculate the classification error rate of the individual and the ratio of the number of selected features to the total number of features. The classification error rate and the ratio of the number of selected features to the total number of features are weighted and fused to obtain the fitness value of each solution.

[0010] Preferably, the binary vector is a binary vector containing only 0 and 1, where 1 represents that the feature is selected and 0 represents that it is not selected.

[0011] Preferably, the process further includes standardizing the medical feature dataset to be processed, including: Limit all feature values ​​in the medical feature dataset to be processed to Within the interval, the medical feature dataset to be processed is divided into training and test sets through ten-fold cross-validation, and the training set KNN model is used.

[0012] This invention also provides a feature selection system based on an improved slime mold optimization algorithm, comprising: The data acquisition module is used to acquire the medical feature dataset to be processed; The feature selection module is used to generate an initial solution set based on the medical feature dataset and iteratively find the optimal feature subset using an improved slime mold optimization algorithm: K-means clustering is used to divide the population in the current initial solution set into several independent clusters, and based on the cluster center vector of the cluster to which each individual in the initial solution set belongs and the difference between the cluster center vectors of other clusters, the individual in the initial solution set is guided to perform directional mutation update based on the current individual position when searching in the feature space, so as to obtain multiple candidate solutions; For multiple candidate solutions, obtain the global population diversity state and global stagnation time of all individuals in the population, as well as the instantaneous position update validity and cumulative stagnation history of each individual; dynamically obtain a crossover probability based on the global population diversity state and global stagnation time, and combine the instantaneous position update validity and cumulative stagnation history of each individual to adaptively select a strategy from a variety of preset update strategies to update the individual positions in multiple candidate solutions, thus obtaining the final multiple candidate solutions; For the final multiple candidate solutions, obtain the fitness value of each solution, aim to minimize the fitness value of each solution, and repeat the search until the maximum number of iterations is reached to obtain the optimal feature subset corresponding to the optimal solution.

[0013] This invention also provides an electronic device, including a memory and a 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 a feature selection method based on an improved slime mold optimization algorithm as described above.

[0014] This invention also provides a computer-readable storage medium for storing a computer program, which, when executed by a processor, implements the steps of a feature selection method based on an improved slime mold optimization algorithm as described above.

[0015] This invention provides a feature selection method and system based on an improved slime mold optimization algorithm. Compared with the prior art, its advantages are as follows: This invention constructs an improved slime mold optimization algorithm, bCMDSMA, by updating individual positions using a K-means clustering-center-based mutation guidance strategy (CIM) and a multi-state history-driven dynamic crossover strategy (MSHDC). This improved algorithm first divides the population into multiple differentiated subgroups through K-means clustering. Then, it uses the cluster centers of each subgroup to guide individual mutations. The cluster centers naturally represent "anchor points" for different potential regions in the search space. This cluster-center-guided mutation operation forces the search to proceed simultaneously across multiple identified advantageous regions, rather than just around a single global optimum. This "multi-anchor point parallel search" mechanism ensures that the search can achieve simultaneous analysis of different potential regions within a broad, high-dimensional solution space. The wide coverage of potential regions enables the search for extremely critical feature combinations to avoid "local optima." Then, based on the crossover probability dynamically calculated from the global population diversity state and global stagnation time, combined with a two-dimensional individual historical state archive containing "instantaneous position update effectiveness" and "cumulative stagnation history," a refined characterization of the search state of each individual is achieved. This "personalized" individual-level differentiated fine-tuning allows the algorithm to have individuals in different states perform different types of search tasks at the same iteration time—some are responsible for local fine mining, some for mid-range exploration, and some for long-range jump exploration. This enables the search for subtle difference feature combinations that are crucial to clinical diagnosis, and ultimately effectively selects the most accurate key features from high-dimensional medical data. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall process provided for an embodiment of the present invention; Figure 2 This is a schematic diagram showing the comparison of clustering effects on a dataset before and after, provided in an embodiment of the present invention. Figure 3 A schematic diagram of the curve showing how the crossover probability changes with the number of iterations, provided for an embodiment of the present invention; Figure 4 This is a schematic diagram showing the convergence curve results of various algorithms on 12 high-dimensional public datasets provided in the embodiments of the present invention. Detailed Implementation

[0017] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0018] Currently, traditional machine learning algorithms often face performance bottlenecks in high-dimensional search spaces and are prone to getting stuck in local optima. Therefore, feature selection has become an indispensable data preprocessing step in machine learning, pattern recognition, and data mining. It can not only significantly reduce the cost of subsequent model training and simplify model structure, but also effectively alleviate overfitting and improve model interpretability. Its core objective can be described as: to select the most informative feature subset while maintaining or even improving the prediction accuracy of the original model. However, traditional feature selection methods often perform poorly when dealing with high-dimensional medical data. Therefore, more and more scholars are focusing on using metaheuristic algorithms to identify the optimal feature subset. However, traditional metaheuristic algorithms still have obvious limitations: they are prone to getting stuck in local optima, leading to the omission of key features, and insufficient convergence accuracy affects clinical discriminative power. Slime Mould optimization algorithm (Slime Mould) Swarm optimization algorithm (SMA) is a swarm intelligence optimization algorithm that performs well in handling feature selection problems. However, when faced with high-dimensional datasets, it also encounters the aforementioned problems, such as getting stuck in local optima and missing key features, and poor convergence accuracy. Therefore, how to avoid the algorithm getting stuck in local optima and missing key features is an important problem that urgently needs to be solved.

[0019] This invention aims to address the problem of SMA (Single-means Association) easily getting trapped in local optima. It introduces a Clustering-Informed Mutation (CIM) strategy based on K-means cluster centers and a Multi-State History-Driven Dynamic Crossover (MSHDC) strategy to update individual positions, enhancing the exploration capability of SMA and reducing the probability of getting trapped in local optima. Furthermore, to enable the improved SMA to handle discrete feature selection problems, it is transformed into a binary algorithm through a transformation function. This invention names the improved algorithm bCMDSMA. Figure 1 As shown, the main steps include: Step 1: Read a high-dimensional medical dataset containing gene expression profiles or biomarker features of specific diseases such as brain tumors, leukemia, or prostate cancer in a specific clinical diagnostic scenario, and perform normalization processing on the medical feature dataset.

[0020] Step 2: Combine all feature values ​​from the high-dimensional medical feature dataset containing specific disease characteristics... The dataset was divided into training and test sets using 10-fold cross-validation.

[0021] Step 3: Initialize the dimensions based on the total number of features in the specific disease medical feature dataset mentioned above. Maximum number of iterations Optimal vector set .

[0022] Step 4: Initialize the algorithm population.

[0023] Step 5: Train the KNN disease diagnosis classification model using the training set above, and calculate the fitness value of each individual (i.e., each feature combination scheme) to evaluate the ability of the feature combination to classify specific diseases.

[0024] Calculating the fitness value of an individual using the KNN model requires converting each individual vector into a binary vector using a transformation function before calculating its fitness value. The specific calculation is as follows:

[0025] .

[0026] .

[0027] .

[0028] .

[0029] in: T The function is a transformation function; Represents the hyperbolic tangent function; Indicates the first t In the next iteration, the individual i of j The numerical value of the dimension. yes A random number that is uniformly distributed within a range; It is the classification accuracy of the KNN model; This represents the fitness value of an individual. and This indicates the preset weight values, set to 0.95 and 0.05; Indicates the number of features selected in the feature subset; This represents the total number of characteristics.

[0030] Step 6: Use bCMDSMA to find the optimal feature subset; specifically including: (1) Use K-means clustering to divide the population into several independent clusters and update the individual position (CIM) according to the cluster center vector.

[0031] CIM uses clustering to divide the population into several independent clusters and updates the individual positions based on the cluster center vectors, enabling the potential dominant regions of the population to be more fully and continuously developed. Its mathematical model is expressed as: .

[0032] .

[0033] in: Indicates the probability of dynamic mutation; Indicates the current iteration number; Indicates the maximum number of iterations; F This is the scaling factor; and Representing an interval A random variable that is uniformly distributed within the range; This represents a random value sampled from a standard normal distribution; Indicates the first t In the next iteration, the individual i No. j The value of the dimension; Indicates the first t In the next iteration, the individual i The cluster center to which it belongs j The value of the dimension; Indicates the first t In the next iteration, the cluster centroid is randomly selected from other subpopulations. j The value of the dimension.

[0034] (2) Obtain the instantaneous location update validity and cumulative stagnation history of each individual.

[0035] An improved greedy selection method is used to obtain the instantaneous position update effectiveness and cumulative stagnation history of each individual, with the main purpose of enriching population diversity. The specific calculation model is as follows:

[0036] .

[0037] .

[0038] .

[0039] in: Represent the objective function to be minimized; Represents an individual i In the t The position at the next iteration; This represents the candidate solutions generated by the evolution operator; It is a binary flag indicating the validity of the instantaneous position update; This indicates the cumulative stagnation history of an individual.

[0040] (3) Use MSHDC to select an appropriate strategy to update the individual.

[0041] MSHDC is used to dynamically adjust the crossover probability of operators based on the population state, and an appropriate strategy is selected to update individuals by integrating the effectiveness of each individual's instantaneous position update and cumulative stagnation history. The specific calculation model is as follows:

[0042] First, the crossover probability of the operator is obtained based on the population state, expressed as: .

[0043] .

[0044] .

[0045] in: This represents the crossover probability calculated based on the current population diversity and global stagnation time, with its parameter settings... , ; This represents a global stall counter with a threshold value. ; Indicates the average distance; Indicates population size; Indicate the dimensions of the problem; Indicates individuals in the current population i With individuals j The Euclidean distance between them, of which and Each represents an individual i and individuals j In the k Values ​​in the dimension.

[0046] Finally, the optimal strategy for updating individuals is selected based on the effectiveness of the individual's instantaneous position update and its cumulative stagnation history, expressed as: .

[0047] .

[0048] in: Indicates the current iteration number; and Indicates the range of values ​​within Random numbers; variables m Indicates length is D The binary sequence is expressed by the mathematical expression: ; , and Indicates the range of values ​​within Random numbers; This represents the best individual in the current population; Indicates individuals in the current population i The cluster center to which it belongs j The value of the dimension; express and The average value between, This represents the maximum threshold at which an individual can be allowed to stagnate.

[0049] Step 7: Determine if the current iteration count exceeds the maximum iteration count. If not, repeat steps 5 and 6.

[0050] Step 8: Output the optimal feature subset, which is the combination of key biomarkers selected from high-dimensional medical data that has the highest clinical differential diagnostic value for the specific disease (such as brain tumors).

[0051] like Figure 2 As shown in the figure, the application effect of the algorithm of the present invention in data clustering analysis is verified by comparison. Figure 2 (a) is a two-dimensional scatter plot of the original data. The data points in the plot are intertwined and do not show clear category boundaries. Figure 2 (b) is a clustering result diagram after K-means clustering based on the CIM strategy of this invention; Figure 2 (b) As can be seen, the data is clearly divided into three categories: cluster 1, cluster 2, and cluster 3, and identified by dots, squares, and triangles respectively; the black pentagrams in the figure represent the cluster centers of each cluster; in comparison... Figure 2 As can be seen from (a) and (b), after processing by the method of the present invention, key features can be effectively extracted from high-dimensional data and noise interference can be reduced, making the originally messy data present significant clustering features; each cluster is compact and the distance between clusters is clear, which proves that the present invention has excellent feature selection performance and data partitioning accuracy when processing complex datasets.

[0052] like Figure 3 As shown, the curves illustrating the change in probability value L with the number of iterations during feature selection are illustrated. The horizontal axis represents the number of iterations, and the vertical axis represents the probability value. In the early stages of the algorithm, the probability value fluctuates wildly, ranging from 0 to 0.8. This indicates that the algorithm performs a large-scale global exploration within the search space to avoid getting trapped in local optima. As the number of iterations increases, the probability value rapidly increases in the later stages and eventually stabilizes around 1.0. This demonstrates that under the optimization mechanism employed in this invention, the algorithm can adaptively adjust and enter a stable local development phase in the later stages, performing a refined search for potential optimal feature subsets, thereby ensuring the convergence and stability of the feature selection results.

[0053] In one embodiment, the present invention performs feature selection on 12 publicly available high-dimensional medical datasets. Each algorithm is run independently 10 times with 10x cross-validation. The datasets are divided into 10 parts, one part is randomly selected as the test set, and the remaining parts are used as the training set. The number of individuals (N) for all evaluation algorithms is set to 20, and the maximum number of iterations (Maxiter) is set to 50. Finally, the algorithms are evaluated based on the average fitness value, average error rate, and average feature count. The Wilcoxon signed-rank test is used on the final results. The following tables show the comparison results of fitness value, classification error rate, and number of selected features, respectively. "+ / = / -" indicates that bCMDSMA is better than the algorithm, similar to the algorithm, and worse than the algorithm, respectively. See Tables 1, 2, and 3.

[0054] In comparative experiments, bCMDSMA was compared with several existing binary optimization algorithms, including the binary Moth to a Flame Optimizer (bMFO), the binary Gravity Search Algorithm (bGSA), the binary Particle Swarm Optimizer (bPSO), the binary Antlion Optimizer (bALO), the binary Bat Algorithm (bBA), the binary Salicylicorice Swarm Optimizer (bSSA), and the binary Slime Fungus Optimizer (bSMA). Among these, bSMA, as a basic version of bCMDSMA, does not incorporate the CIM and MSHDC mechanisms. The remaining algorithms are mature methods that have been widely applied and performed excellently in feature selection, and their effectiveness has been verified on multiple datasets. These algorithms were selected as benchmarks to objectively evaluate the performance advantages of this invention and further verify its advancement and practicality in feature selection tasks.

[0055] Table 1. Comparison of average fitness values ​​of bCMDSMA with other algorithms on 12 high-dimensional datasets. Table 2 Comparison of average classification error rates of bCMDSMA and other algorithms on 12 high-dimensional datasets. Table 3 Comparison of average feature selection number of bCMDSMA and other algorithms on 12 high-dimensional datasets. Figure 4 The results show the convergence curves of bCMDSMA and various algorithms on 12 high-dimensional medical public datasets.

[0056] This invention employs a collaborative mechanism of "CIM multi-anchor parallel search" and "MSHDC individual differentiated regulation." The CIM strategy ensures the parallel maintenance and exploration of multiple different feature combination patterns, preventing regions containing features with weak initial signals but potential synergistic value from being prematurely abandoned. Simultaneously, the MSHDC strategy, through continuous monitoring of individual states, can promptly identify and disturb individuals stagnating at local extrema, granting them the ability to penetrate search barriers and leap to entirely new solution spaces. The synergy of these two mechanisms is equivalent to simultaneously deploying multiple search formations distributed across different regions within a vast high-dimensional biomarker search space, with each member possessing the ability to proactively adjust its search range and direction based on its own state. This enables the algorithm to effectively discover key biomarkers that require specific feature combinations to demonstrate their value, significantly reducing the risk of misdiagnosis and missed diagnosis in clinical practice due to feature omissions.

[0057] The CIM strategy of this invention makes it less likely for the algorithm to miss any potential feature combination regions, while the MSHDC strategy ensures that a thorough and detailed search can be performed in each potential region through adaptive adjustment of individual states, avoiding premature satisfaction with suboptimal solutions. The optimal feature subset produced by this high-quality search process not only eliminates redundant and noisy features, but more importantly, retains those key feature combinations that have differential diagnostic value in complex clinical classifications.

[0058] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A feature selection method based on an improved slime optimization algorithm, characterized in that, Includes the following steps: Obtain the medical feature dataset to be processed; An initial solution set is generated based on a medical feature dataset, and an improved slime mold optimization algorithm is used to iteratively find the optimal feature subset: K-means clustering is used to divide the population in the current initial solution set into several independent clusters, and based on the cluster center vector of the cluster to which each individual in the initial solution set belongs and the difference between the cluster center vectors of other clusters, the individuals in the initial solution set are guided to perform directional mutation updates based on the current individual position when searching in the feature space, thereby obtaining multiple candidate solutions; For multiple candidate solutions, obtain the global population diversity state and global stagnation time of all individuals in the population, as well as the instantaneous position update validity and cumulative stagnation history of each individual; dynamically obtain a crossover probability based on the global population diversity state and global stagnation time, and combine the instantaneous position update validity and cumulative stagnation history of each individual to adaptively select a strategy from a variety of preset update strategies to update the individual positions in multiple candidate solutions, thus obtaining the final multiple candidate solutions; For the final multiple candidate solutions, obtain the fitness value of each solution, aim to minimize the fitness value of each solution, and repeat the search until the maximum number of iterations is reached to obtain the optimal feature subset corresponding to the optimal solution.

2. The feature selection method based on the improved slime mold optimization algorithm according to claim 1, characterized in that, The targeted mutation update includes: K-means clustering is used to divide the population in the current initial solution set into several independent clusters, and the individual position is mutated and updated based on the difference between the cluster center vector of the cluster to which each individual belongs in the initial solution set and the cluster center vector of other clusters. When guiding individual locations to mutate and update, a dynamic mutation probability is set based on the mutation guidance strategy CIM, which linearly decreases from a higher value to a lower value as the iteration progresses. When the number of random variables is greater than the preset first dynamic mutation probability, the difference vector between the cluster center of the individual's current cluster and the cluster center of another randomly selected cluster is used to guide cross-regional directional mutation of the individual. When the number of random variables is greater than the preset second dynamic mutation probability but less than the preset first dynamic mutation probability, a Gaussian random perturbation is superimposed on the cluster center of the individual, and exploratory updates are performed at the cluster center.

3. The feature selection method based on the improved slime mold optimization algorithm according to claim 1, characterized in that, The acquisition of the instantaneous position update validity and cumulative stagnation history for each individual includes: The greedy selection mechanism is used to obtain the instantaneous position update effectiveness and cumulative stagnation history of each individual in the population of multiple candidate solutions; When using the greedy selection mechanism, if the fitness value of a candidate solution is higher than the preset fitness value, the solution is accepted and used as the new position. At the same time, the instantaneous update validity of the individual is marked as successful, and the accumulated stagnation count is cleared to zero. If the fitness value of a candidate solution is lower than the preset fitness value, the solution is rejected and the original position is maintained. The instantaneous update validity is marked as failed, and the accumulated stagnation count is incremented by one.

4. The feature selection method based on the improved slime mold optimization algorithm according to claim 3, characterized in that, The acquisition of the final multiple candidate solutions includes: Based on the global population diversity state and global stagnation time, the crossover probability of the operator is obtained. The crossover probability of the operator is jointly determined by the diversity level represented by the average inter-individual distance of the population and the global stagnation time represented by the number of consecutive generations without updating the global optimal solution. Based on the crossover probability of the operator, and combined with the instantaneous position update effectiveness and cumulative stagnation history of the individual, an adaptive strategy is selected from a variety of preset update strategies to update the individual, thus obtaining multiple candidate solutions. The update strategy includes: for successful and non-stagnant individuals, an open-ended update is performed using the difference between the composite guiding vector and the current individual; for unsuccessful but non-stagnant individuals, a local exploration is performed using the difference between the random individual and the current individual; for successful but stagnant individuals, a guided jump exploration is performed using the difference between the random individual and the composite guiding vector; and for unsuccessful and stagnant individuals, a strongly random perturbation exploration is performed using the difference between two random individuals.

5. The feature selection method based on the improved slime mold optimization algorithm according to claim 1, characterized in that, The process of obtaining the fitness value for each solution includes: Each individual vector within the final multiple candidate solutions is converted into a binary vector using a transformation function. The KNN model is then used to calculate the classification error rate of the individual and the ratio of the number of selected features to the total number of features. The classification error rate and the ratio of the number of selected features to the total number of features are weighted and fused to obtain the fitness value of each solution.

6. The feature selection method based on the improved slime mold optimization algorithm according to claim 5, characterized in that, The binary vector is a binary vector containing only 0 and 1, where 1 represents that the feature is selected and 0 represents that it is not selected.

7. A feature selection method based on an improved slime mold optimization algorithm according to claim 5, characterized in that, It also includes standardization processing of the medical feature dataset to be processed, including: Limit all feature values ​​in the medical feature dataset to be processed to Within the interval, the medical feature dataset to be processed is divided into training and test sets through ten-fold cross-validation, and the training set KNN model is used.

8. A feature selection system based on an improved slime mold optimization algorithm, characterized in that, include: The data acquisition module is used to acquire the medical feature dataset to be processed; The feature selection module is used to generate an initial solution set based on the medical feature dataset and iteratively find the optimal feature subset using an improved slime mold optimization algorithm: K-means clustering is used to divide the population in the current initial solution set into several independent clusters, and based on the cluster center vector of the cluster to which each individual in the initial solution set belongs and the difference between the cluster center vectors of other clusters, the individual in the initial solution set is guided to perform directional mutation update based on the current individual position when searching in the feature space, so as to obtain multiple candidate solutions; For multiple candidate solutions, obtain the global population diversity state and global stagnation time of all individuals in the population, as well as the instantaneous position update validity and cumulative stagnation history of each individual; dynamically obtain a crossover probability based on the global population diversity state and global stagnation time, and combine the instantaneous position update validity and cumulative stagnation history of each individual to adaptively select a strategy from a variety of preset update strategies to update the individual positions in multiple candidate solutions, thus obtaining the final multiple candidate solutions; For the final multiple candidate solutions, obtain the fitness value of each solution, aim to minimize the fitness value of each solution, and repeat the search until the maximum number of iterations is reached to obtain the optimal feature subset corresponding to the optimal solution.

9. 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 based on the improved slime mold optimization algorithm as described in any one of claims 1 to 7.

10. 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 based on an improved slime mold optimization algorithm as described in any one of claims 1 to 7.

Images