XGBoost innovation capability evaluation method and system based on improved particle swarm optimization parameter adjustment

Abstract

The invention discloses an XGBoost innovation capability evaluation method and system based on improved particle swarm optimization parameter adjustment, and the method comprises the steps: collecting multi-dimensional index data of a target object, and carrying out the preprocessing of the collected multi-dimensional index data; performing standardization processing and principal component extraction on scale scores and behavior numerical value characteristics by adopting principal component analysis; performing global search and optimization on the target hyper-parameters by using an improved particle swarm optimization algorithm by taking XGBoost as a base learner; and training an XGBoost model by using the optimized hyper-parameters, and inputting new sample data to predict the innovation ability of the target object to obtain a prediction result, the prediction result comprising an innovation ability score or grade classification.

Description

Technical Field

[0001] This application relates to the field of machine learning technology, and more specifically, to an XGBoost innovation capability evaluation method and system based on improved particle swarm optimization parameter tuning. Background Technology

[0002] Currently, the methods for assessing college students' innovative abilities mainly fall into the following categories: The first category is the scale-based assessment method. A common approach is to design or introduce existing psychological and educational scales, such as the Torrance Test of Creative Thinking (TTCT), the Creativity Self-Efficacy Scale, and the Learning Motivation Scale. This type of method has advantages such as ease of implementation and promotion, but it also has significant drawbacks. These mainly manifest as over-reliance on student self-reporting and susceptibility to influences such as social expectation effects and motivational responses, leading to biased assessment results.

[0003] The second method is expert review. This involves teachers, experts, or project mentors scoring or rating students' innovative abilities. While this method incorporates practical experience, the lack of standardized quantitative criteria leads to significant subjectivity and inconsistency in the evaluation results, making it unsuitable for large-scale systematic evaluations.

[0004] The third type is the outcome-oriented approach. This method assesses innovation ability based on students' explicit achievements such as competition results, patents, papers, and entrepreneurial projects. While emphasizing results and reflecting the level of innovation output to some extent, it neglects the process and potential of innovation ability. For example, some students may possess strong innovation potential but fail to produce explicit results due to a lack of resources and opportunities, thus being underestimated under this type of evaluation system.

[0005] In recent years, with the rapid development of artificial intelligence and machine learning, some studies have attempted to apply data-driven modeling methods to assess the innovation capabilities of university students. For example, methods such as random forests, support vector machines, and neural networks are used to model relevant student data. These methods can capture nonlinear relationships and improve prediction performance, but they also have certain limitations: First, the parameter tuning process is complex, and without effective optimization strategies, the model is prone to overfitting or underfitting; second, while some methods, such as deep neural networks, have high prediction accuracy, their interpretability is insufficient, hindering their widespread application in educational settings; and third, existing research largely remains at the theoretical verification stage, lacking a systematic application framework. Summary of the Invention

[0006] To address at least one deficiency or improvement need in existing technologies, this invention provides an XGBoost innovation ability evaluation method and system based on improved particle swarm optimization parameter tuning. It aims to solve problems such as strong subjectivity, insufficient data dimensions, limited model parameter optimization capabilities, and poor predictive interpretability in existing methods for evaluating university students' innovation abilities. By combining intelligent optimization algorithms with machine learning models, this invention can achieve high-precision, interpretable, and generalizable innovation ability evaluation in a multi-dimensional data environment, providing scientific support for talent cultivation and educational decision-making in universities.

[0007] To achieve the above objectives, according to a first aspect of the present invention, an XGBoost innovation capability evaluation method based on improved particle swarm optimization is provided. The method includes: collecting multidimensional index data of a target object; preprocessing the collected multidimensional index data; standardizing and extracting principal components from scale scores and behavioral numerical features using principal component analysis; using XGBoost as a base learner and employing an improved particle swarm optimization algorithm to globally search and optimize the target hyperparameters; training an XGBoost model using the optimized hyperparameters; and predicting the innovation capability of the target object using new sample data to obtain a prediction result, wherein the prediction result includes an innovation capability score or a classification level.

[0008] In an exemplary embodiment, the standardization and principal component extraction of scale scores and behavioral numerical features using principal component analysis includes: constructing a correlation coefficient matrix and calculating eigenvalues ​​and eigenvectors; drawing a scree plot and determining the number of principal components to be extracted based on the inflection point of the eigenvalue decline; calculating the cumulative explained variance and determining the number of principal components using the cumulative explained variance and the scree plot; generating a principal component loading matrix and performing semantic interpretation on each principal component in conjunction with the variable loading values.

[0009] In an exemplary embodiment, the target hyperparameters include: the learning rate, number of trees, maximum depth, row / column sampling ratio, minimum sample weight of leaf nodes, and regularization coefficient of the XGBoost model, all of which are automatically optimized within a preset range using the IPSO algorithm.

[0010] In one exemplary embodiment, the method further includes: performing interpretive analysis on the prediction results using SHAP values; calculating the contribution of each feature to the prediction results; and generating a ranking and visualization report of the factors influencing innovation capability.

[0011] In an exemplary embodiment, the step of using XGBoost as a base learner and employing an improved particle swarm optimization algorithm to perform a global search and optimization of the target hyperparameters includes: progressively stacking multiple weak learners to form a strong learner, designing the objective function of XGBoost; performing a second-order gradient approximation, and determining the optimal split by calculating the split gain.

[0012] In an exemplary embodiment, the step of using XGBoost as the base learner and employing an improved particle swarm optimization algorithm to perform a global search and optimization of the target hyperparameters includes: adjusting the inertia weights using a cosine curve adaptive strategy; and employing an adaptive learning factor adjustment mechanism to make the group learning factor and the individual learning factor change dynamically with the iteration process.

[0013] In an exemplary embodiment, before predicting the innovation capability of the target object using the input new sample data to obtain the prediction result, the method further includes: defining a particle representation and a fitness function, wherein each particle represents a candidate solution vector containing the target hyperparameters of XGBoost; performing an iterative search using an improved particle swarm optimization algorithm, the improved particle swarm optimization algorithm including adaptive inertia weights and dynamic learning factors; and when the iteration converges, obtaining the optimal particle solution vector and decoding to obtain the optimal parameter set and the optimal feature set.

[0014] According to a second aspect of the present invention, an XGBoost innovation capability evaluation system based on improved particle swarm optimization parameter tuning is also provided, which executes the above-described XGBoost innovation capability evaluation method based on improved particle swarm optimization parameter tuning, and includes: a data acquisition module, a data preprocessing module, a model prediction module, and a result visualization module.

[0015] According to a third aspect of the present invention, a computer-readable storage medium is also provided, wherein a computer program is stored therein, wherein the computer program is configured to execute the above-described XGBoost innovation capability evaluation method based on improved particle swarm optimization parameter tuning at runtime.

[0016] According to a fourth aspect of the present invention, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the above-described XGBoost innovation capability evaluation method based on improved particle swarm optimization parameter tuning via the computer program.

[0017] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: (1) This invention provides an XGBoost innovation capability evaluation method based on improved particle swarm optimization parameter tuning, which integrates multi-source data and can comprehensively reflect the complexity and multidimensionality of innovation capability, avoiding the one-sidedness of single-dimensional evaluation; second, the key parameters of XGBoost are optimized by using the IPSO algorithm, which significantly improves the prediction performance and robustness of the model and effectively overcomes the shortcomings of traditional parameter tuning methods; third, the output results have good interpretability, and the key factors affecting innovation capability can be revealed by feature importance analysis and SHAP value method, thereby enhancing the guiding value of the application; fourth, the system architecture is complete and has strong scalability and extensibility.

[0018] (2) It realizes the intelligent, precise and interpretable assessment of college students' innovation ability, providing important support for the reform of innovation education and the optimization of talent training models in universities, and has outstanding theoretical significance and application prospects. It can be widely used in efficient educational practices and can be extended to the ability assessment scenarios of other groups. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating an optional XGBoost innovation capability evaluation method based on improved particle swarm optimization parameter tuning, provided for embodiments of this application; Figure 2 This application provides an optional data acquisition process diagram. Figure 3 A schematic diagram of an optional improved particle swarm optimization algorithm provided for an embodiment of this application; Figure 4 A schematic diagram of an optional XGBoost algorithm provided for an embodiment of this application; Figure 5 A schematic diagram of another optional XGBoost innovation capability evaluation model based on improved particle swarm optimization parameter tuning provided for embodiments of this application; Figure 6 A schematic diagram of an optional XGBoost innovation capability evaluation system based on improved particle swarm optimization parameter tuning provided for an embodiment of this application; Figure 7 This is a schematic diagram of an optional electronic device provided in an embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0022] The terms "first," "second," "third," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0023] According to one aspect of the embodiments of this application, an XGBoost innovation capability evaluation method based on improved particle swarm optimization parameter tuning is provided. The following is in conjunction with... Figure 1 This application describes the XGBoost innovation capability evaluation method based on improved particle swarm optimization parameter tuning provided in the embodiments of this application.

[0024] Figure 1 This is a flowchart illustrating an optional XGBoost innovation capability evaluation method based on improved particle swarm optimization parameter tuning, as provided in an embodiment of this application. Figure 1 As shown, the process of this method may include the following steps: S102, Collect multi-dimensional indicator data of the target object, and preprocess the collected multi-dimensional indicator data; S104, Principal component analysis was used to standardize the scale scores and behavioral numerical features and extract principal components. S106 uses XGBoost as the base learner and employs an improved particle swarm optimization algorithm to perform global search and optimization of the target hyperparameters. S108, using the optimized hyperparameters to train the XGBoost model, inputting new sample data to predict the innovation ability of the target object and obtaining the prediction result, the prediction result including innovation ability score or level classification.

[0025] This application provides an innovative ability evaluation method based on improved particle swarm optimization (IPSO) parameter tuning, namely XGBoost (eXtreme Gradient Boosting), which is applicable to efficient educational practices and can be extended to ability evaluation scenarios for other groups.

[0026] To overcome the shortcomings of traditional methods, this invention establishes a scientific, objective, and intelligent evaluation system for college students' innovation capabilities. Through multi-source data collection and fusion, including multi-dimensional indicators such as psychological traits, behavioral data, and environmental resources, a comprehensive indicator system is established, thus avoiding reliance solely on subjective evaluation. Meanwhile, while the traditional XGBoost model possesses powerful nonlinear fitting capabilities, its parameter tuning heavily depends on experience and manual search, easily leading to problems such as limited search range and getting trapped in local optima. This invention introduces an improved particle swarm optimization algorithm to adaptively optimize key hyperparameters globally, thereby improving the model's predictive performance.

[0027] Optionally, combined Figure 1 and Figure 4 As shown, the process of this method may include the following steps: Step 1: Data preprocessing. (1) Data cleaning: Delete abnormal data, and impute a small number of random missing values ​​according to the mean / median; (2) Scale synthesis: Calculate the mean / total score for each scale (such as openness). If multiple scales are used to synthesize a comprehensive factor, clarify the weighting rules and record them; (3) Feature encoding: Encode all type variables (grade, major type). Use one-hot encoding for categories with small cardinality and target encoding or grouping and merging for categories with large cardinality. Binarization (yes / no) is mapped to 0 / 1; (3) Feature construction: Select the Min-Max normalization algorithm to normalize all feature data and map the feature data to a fixed range (such as [-1,1]) to ensure that the data distribution is uniform; The formula is: In the formula, The data is after normalization; This represents the original sample data before normalization; This is represented as the minimum value of the feature values ​​in the original data; The maximum value of the original data feature is represented as (4) Divide the dataset: According to the size of the data and the preset standard, the normalized data is randomly divided into training set, validation set and test set.

[0028] like Figure 2As shown, in the data collection phase, this study obtained relevant information about individual students by combining questionnaires with behavioral data collection. The questionnaire covered basic demographic characteristics, personality traits (such as openness, growth mindset, and intrinsic motivation), innovative behaviors (such as the frequency of generating new ideas, participation in competitions and research projects), utilization of learning and research resources, and self-assessment of abilities, using a combination of Likert scales and behavioral counting questions. Regarding sample size, the study collected complete questionnaires from hundreds to thousands of university students, and combined this with some learning behavior data (such as participation in academic competitions and research projects) to form the initial dataset.

[0029] Step Two: Principal Component Analysis. The scale scores and behavioral numerical features were standardized, and principal components were extracted for exploratory analysis and visualization. The number of principal components was determined using the cumulative explained variance and scree plot, and the actual meaning of each principal component was interpreted in conjunction with the loading matrix. This process helps to discover potential clustering patterns in the innovation ability characteristics of university students, and principal components can also be used as input variables after dimensionality reduction when necessary.

[0030] Step 3: Model Construction. In the Python environment, the improved particle swarm optimization algorithm is implemented using the pyswarms library, and the XGBoost model is initialized using the xgboost library. Using XGBoost as the base learner, the improved particle swarm optimization algorithm automatically searches for key hyperparameters of XGBoost and optionally implements feature selection, thereby obtaining a model with optimal generalization performance under cross-validation and as minimal as possible.

[0031] Step 4: Model Prediction. (1) Input Data Preprocessing: Ensure that the feature dimensions and order of the new samples are consistent with those of the training, and that the missing value filling method is the same. (2) Prediction Output: Regression Task: Output predicted values, such as innovation ability scores; Classification Task: Output probability P(y=1|x), and give a level judgment based on the threshold or ranking. (3) Interpretive Analysis: Calculate the contribution of each feature to the prediction result using SHAP values. The formula is as follows, where φj represents the contribution of the j-th feature: (4) Uncertainty estimation: The predicted mean and variance are calculated through model ensemble (multiple training) to give the interval estimate.

[0032] Step 5: Model Evaluation. Evaluation metrics such as Mean Absolute Error (MSE), Precision, and F1-score are used to assess the model's performance.

[0033] This embodiment collects multidimensional indicator data of the target object and preprocesses the collected multidimensional indicator data; principal component analysis is used to standardize the scale scores and behavioral numerical features and extract principal components; XGBoost is used as the base learner, and an improved particle swarm optimization algorithm is used to perform global search and optimization of the target hyperparameters; the optimized hyperparameters are used to train the XGBoost model, and new sample data is input to predict the innovation ability of the target object to obtain the prediction result, which includes an innovation ability score or level classification. By combining intelligent optimization algorithms with machine learning models, this invention can achieve high-precision, interpretable, and generalizable innovation ability assessment in a multidimensional data environment, providing scientific support for talent cultivation and educational decision-making in universities.

[0034] In an exemplary embodiment, the standardization and principal component extraction of scale scores and behavioral numerical features using principal component analysis includes: Construct the correlation coefficient matrix and calculate the eigenvalues ​​and eigenvectors; Draw a scree plot and determine the number of principal components to be extracted based on the inflection point of eigenvalue descent. Calculate the cumulative explained variance, and determine the number of principal components using the cumulative explained variance and scree plot; Generate principal component loading matrices and interpret the semantics of each principal component by combining the variable loading values.

[0035] In an exemplary embodiment, the target hyperparameters include: the learning rate, number of trees, maximum depth, row / column sampling ratio, minimum sample weight of leaf nodes, and regularization coefficient of the XGBoost model, all of which are automatically optimized within a preset range using the IPSO algorithm.

[0036] In one exemplary embodiment, the method further includes: The prediction results are interpreted using SHAP values; Calculate the contribution of each feature to the prediction results and generate a ranking and visualization report of the factors influencing innovation capability.

[0037] In one exemplary embodiment, the step of using XGBoost as a base learner and employing an improved particle swarm optimization algorithm to perform a global search and optimization of the target hyperparameters includes: The objective function of XGBoost is designed by progressively stacking multiple weak learners to form a strong learner. A second-order gradient approximation is performed, and the optimal split is determined by calculating the split gain.

[0038] In this application, for Figure 4This section provides a detailed introduction to the principles of the XGBoost algorithm. XGBoost (Extreme Gradient Boosting) is a highly efficient ensemble learning algorithm based on Gradient Boosting Decision Tree (GBDT). It iteratively builds new trees to correct the residuals of existing models, thereby gradually improving prediction performance. Its core idea is: Model construction method: An additive model is used, in which multiple weak learners (CART regression trees) are progressively stacked to form a strong learner. Let F represent the predicted value in the t-th iteration, and let F represent the set of regression trees.

[0039] Objective function design: The objective function of XGBoost consists of a loss function and a regularization term, where... This represents the training error, such as squared loss or logarithmic loss. This represents the regularization term, which controls the complexity of the tree and prevents overfitting.

[0040] Second-order gradient approximation: Unlike traditional GBDT which uses the first-order gradient, XGBoost introduces a second-order Taylor expansion while utilizing the first-order gradient g. i and second-order h i Improve and optimize efficiency and accuracy:

[0041] Node splitting criterion: The optimal split is determined by calculating the splitting gain (Gain), where G... L H L Let G represent the sum of the gradients of the left subtree and the sum of the second-order gradients. R H R Let γ represent the sum of the gradients of the right subtree and the sum of the second-order gradients, and let γ represent the penalty function for node splitting.

[0042] XGBoost's performance largely depends on hyperparameters (such as tree depth max_depth, learning rate eta, subsample rate, feature sampling rate colsample_bytree, number of trees n_estimators, etc.). Manual parameter tuning or grid search is inefficient, and random search is unstable. Therefore, IPSO is introduced to globally optimize hyperparameters.

[0043] In one exemplary embodiment, the step of using XGBoost as a base learner and employing an improved particle swarm optimization algorithm to perform a global search and optimization of the target hyperparameters includes: An adaptive strategy based on cosine curves is used to adjust the inertia weights; An adaptive learning factor adjustment mechanism is adopted to make the group learning factor and individual learning factor change dynamically with the iteration process.

[0044] In this application, Particle Swarm Optimization (PSO) is a global optimization algorithm based on swarm intelligence, proposed by Kennedy and Eberhart in 1995. It is inspired by the cooperative behavior of flocks of birds or schools of fish in the process of searching for food.

[0045] In PSO, candidate solutions are represented as particles, each with two attributes in the search space: position representing a potential solution and velocity representing the search direction and step size. The algorithm's iterative update rule mainly relies on two aspects: individual experience: the best position a particle has historically found; and swarm experience: the globally best position found by all particles in the swarm. The update formula is:

[0046] in, and These are the t-th iteration and the 1st iteration. i The position and velocity vector of each particle It is the best location in particle history. It is the best position in the entire system. w Inertial weighting is used to balance global search and local development; , The learning factor controls the influence of individuals and groups on particles, respectively. , This represents a random number in the range [0,1]. By continuously updating the velocity and position, the particle swarm gradually converges to the optimal solution.

[0047] Improved Particle Swarm Optimization Algorithm: The standard particle swarm optimization algorithm has fixed parameters, including inertia weights. w The inertia of particles is described; it is relatively large in the early stages of the algorithm to ensure that each particle flies independently and fully explores the space, and becomes smaller in the later stages to learn more from other particles. Furthermore, , The flight step size of the particle towards its individual optimal position and the global optimal position are controlled separately, in the early stage. Larger, later The parameters are relatively large, thus balancing the global and local search capabilities of particles. These three parameters collectively influence the particle's flight direction, causing the current particle to be unable to quickly reach a better position even if other particles find it, due to its high inertia. The standard particle swarm optimization algorithm only allows information from the globally optimal particle; there is a lack of information sharing between particles, and each update only points to the globally optimal solution, unaffected by other particles. Therefore, particle swarm optimization suffers from poor local search capability, susceptibility to local optima, and low search accuracy. To address these issues, such as... Figure 3 As shown, the present invention has made the following improvements: Dynamic Adjustment of Inertia Weight: In practical applications, particle swarm optimization (PSO) algorithms are prone to getting trapped in local optima. To address this issue, an adaptive adjustment control concept is introduced to improve the inertia weight. A larger inertia weight leads to global exploration, while a smaller inertia weight leads to local exploration, and the current exploration region can be fine-tuned. Furthermore, the current iteration number and the population size during algorithm updates are also considered. The formula for the inertia weight w is as follows:

[0048] in, This represents the maximum inertia weight value. The minimum inertia weight value is represented by t, the current iteration number is represented by t, and the maximum iteration number is represented by T. These are the inertia weight parameters for the particle swarm. w The decrease in parameters as the number of iterations increases is beneficial for particles to move near the target value, thus gradually converging the optimal solution of the algorithm to the target value. Calculate the initial parameters. w A rapid decline in parameters can lead to insufficient early exploration, preventing PSO from finding the global optimum. If the algorithm converges too slowly in later stages, and its later development is insufficient, PSO may fail to accurately search local regions. Based on this phenomenon, the parameters... w The problem can be effectively solved by using a curve adaptive adjustment method, employing an adaptive value method and parameter... w It is expressed as follows:

[0049] Cosine curve adaptive strategy replaces traditional linear adaptive parameter update. w Methods, inertia weights w The choice of particle is closely related to the search range and directly affects the global and local search capabilities: w The larger the value, the stronger the global search capability. Compared to the linear decreasing strategy, the cosine adaptive strategy has a larger parameter value. wThe trends of change differ. Linear decreasing decreases at a constant rate, while cosine decreasing decreases slowly in the early and late stages, and more rapidly in the middle. This characteristic gives particles a stronger global search capability in the early stages and is more conducive to local exploration in the later stages. In addition, the weight value of the cosine strategy is usually greater than that of the linear strategy in the early stages, which helps to expand the search space and enhance the detection capability; while it is less than that of the linear strategy in the later stages, thereby narrowing the search range and improving the convergence accuracy.

[0050] Adaptive Learning Factors: In the standard particle swarm optimization algorithm, the particle's velocity update depends on two learning factors (acceleration constants): c1 represents the individual learning factor, indicating the particle's ability to approach its own historical optimal solution, and c2 represents the swarm learning factor, indicating the particle's ability to approach the swarm's global optimal solution. Generally, fixed values ​​are used, such as c1=c2=2.0. However, fixed values ​​have problems: if c1 and c2 are too large, the particles will oscillate frequently, reducing the convergence speed; if c1 and c2 are too small, the search ability is insufficient, and it is easy to get trapped in local optima; fixed values ​​lack flexibility and cannot balance early global search with later local exploitation. Therefore, this invention adopts an adaptive learning factor adjustment mechanism, making c1 and c2 dynamically change with the iteration process, thereby better balancing search and exploitation.

[0051] A linear decreasing / increasing adaptive adjustment is adopted. Initially, to enhance individual learning ability (larger c1, smaller c2), diverse search methods are encouraged; while in the later stage, to enhance the group learning ability (smaller c1, larger c2), convergence is accelerated. The formula is as follows:

[0052] Where t represents the current iteration number and T represents the maximum iteration number. , , , The parameters are set within a preset range. As iterations proceed, particles gradually reduce their reliance on their own experience and increase their reliance on the collective experience, achieving exploration first and then convergence.

[0053] In summary, this method dynamically adjusts the inertia weights. w The learning factors c1 and c2 enable the algorithm to better perform global search in the early stage and local search in the later stage. At the same time, it can increase the diversity of the population, improve the search and optimization ability of particles, avoid particles getting trapped in local optima, and thus improve the accuracy and convergence speed of the algorithm.

[0054] In one exemplary embodiment, before predicting the innovation capability of the target object using the new input sample data to obtain a prediction result, the method further includes: Define the particle representation and fitness function, where each particle represents a candidate solution vector containing the target hyperparameters of XGBoost; An improved particle swarm optimization algorithm is used for iterative search, wherein the improved particle swarm optimization algorithm includes adaptive inertia weights and dynamic learning factors; When the iteration converges, the optimal particle solution vector is obtained, and the optimal parameter set and the optimal feature set are decoded.

[0055] Optionally, during model building, for example in a Python environment, an improved particle swarm optimization algorithm can be implemented using the pyswarms library, while the XGBoost model can be initialized using the xgboost library. Utilizing XGBoost as the base learner, the improved particle swarm optimization algorithm automatically searches for key XGBoost hyperparameters and optionally implements feature selection, thereby obtaining a model with optimal generalization performance under cross-validation and that is as concise as possible.

[0056] Define particle representation: Each particle represents a candidate solution vector X, containing the key hyperparameters of XGBoost: The meaning and values ​​of the parameters are as follows:

[0057] Fitness function definition: The fitness function measures the performance of candidate solutions on cross-validation and can incorporate a feature count penalty to encourage sparsity. The model focuses on both regression and classification, and can employ a weighted synthesis, where... Indicates the return to the task. Indicates a classification task. RMSE is the root mean square error, and AUC is the area under the ROC curve. The choice should be made based on the task's focus.

[0058] To encourage a small number of features, a penalty term is added to the fitness score, where |x(f)| is the number of selected features, and β≥0 is the penalty function (taken as 0.005-0.05):

[0059] IPSO iterative search: PSO particle velocity / position updated to:

[0060] in, and They are the first i The position and velocity vector of each particle It is the best location in particle history. It is the best position in the entire system. , Let be a random vector in (0,1). It is inertial weight. , This is the learning factor. The improved particle swarm optimization algorithm includes adaptive inertia weights and a dynamic learning factor:

[0061]

[0062] The IPSO parameters are set as follows:

[0063] Training the XGBoost model: After IPSO converges, the optimal particle x* is obtained. The optimal parameter set and feature set are decoded. Then, the XGBoost model is trained on the entire training set using these hyperparameters, and the model parameters are saved. The optimal parameters are, for example:

[0064] Model prediction. (1) Input data preprocessing: Ensure that the feature dimensions and order of the new samples are consistent with those of the training, and that the missing value filling method is the same. (2) Prediction output: Regression task: Output predicted value, such as innovation ability score; Classification task: Output probability P(y=1|x), and give a level judgment based on the threshold or ranking. (3) Interpretive analysis: Calculate the contribution of each feature to the prediction result using SHAP value, as shown in the following formula, where φj represents the contribution of the j-th feature: (4) Uncertainty estimation: The predicted mean and variance are calculated through model ensemble (multiple training) to give the interval estimate.

[0065] Model evaluation uses metrics such as mean absolute error (MSE), accuracy, and F1-score to assess model performance.

[0066] (1) Mean Absolute Error (MSE) measures the mean squared error between the model's predicted probability and the true class. The smaller the MSE, the closer the model's predicted probability is to the true label. Its calculation formula is as follows, where N is the number of samples. yi For real labels, The probability distribution predicted by the model:

[0067] (2) Precision refers to the proportion of samples correctly predicted by the model out of all samples. The calculation formula is as follows, where TP (True Positive): the number of samples correctly predicted as a certain class; FP (False Positive): the number of samples incorrectly predicted as the same class.

[0068] (3) The coefficient of determination, usually denoted as R², measures the extent to which the model explains the variance in the data. When R² = 1: the model makes a perfect prediction; when R² = 0: the model performs as poorly as simply using the mean, offering no explanatory power; when R² < 0: the model performs worse than using the mean. Its formula is as follows, where... yi For the true value, For predicted values, The mean:

[0069] By introducing an improved Particle Swarm Optimization (IPSO) algorithm, adaptive optimization of hyperparameters is achieved. In IPSO, each particle represents a set of XGBoost hyperparameter combinations, and the optimal solution is continuously searched in the parameter space based on the velocity and position update formulas. Unlike the traditional PSO algorithm, IPSO employs an adaptive inertia weight decay mechanism and a dynamic learning factor adjustment strategy, effectively avoiding premature convergence and significantly enhancing global optimization capabilities.

[0070] According to another aspect of the embodiments of this application, an evaluation system for implementing the above-described XGBoost innovation capability evaluation method based on improved particle swarm optimization parameter tuning is also provided. This XGBoost innovation capability evaluation system based on improved particle swarm optimization parameter tuning, such as... Figure 6 As shown, the system includes: a data acquisition module, a data preprocessing module, a model prediction module, and a results visualization module. This system can run on a server or cloud platform, supports real-time processing and batch prediction of large-scale student data, and has good scalability and promotional value.

[0071] The data collection module is designed to collect relevant student data through methods such as questionnaires, covering multiple dimensions of indicators such as psychological traits, learning motivation, research experience, teamwork, and environmental support. The data preprocessing module is designed to standardize the collected data, including data cleaning, noise reduction, missing value imputation, normalization, and feature extraction, ultimately forming a dataset that can be used for modeling, and randomly dividing it into training set, validation set, and test set. The model prediction module is designed to construct a prediction model to predict the innovation ability scores of university student samples and output corresponding level classifications. Simultaneously, it incorporates interpretive methods such as SHAP values ​​to reveal the contribution of each input feature to the individual prediction results. For example, a student's high score may primarily stem from strong learning motivation and frequent participation in research, while a low score may be related to insufficient teamwork or low resource utilization.

[0072] The results visualization module is designed to generate a group-level report on the distribution of innovation capabilities through model analysis and prediction, and to provide individualized evaluation results and improvement suggestions for each student. The group report mainly presents the overall innovation capability level of students in each grade or major, the ranking of the importance of characteristics, and the analysis of influencing factors, which can provide a basis for universities to make decisions on curriculum reform and resource allocation. The individual report is presented in the form of scores, grades, key influencing factors, and improvement suggestions, providing targeted references for students and their instructors.

[0073] Leveraging XGBoost's powerful nonlinear fitting capabilities and IPSO's global hyperparameter optimization, this method outperforms traditional statistical methods and unoptimized machine learning models in terms of prediction accuracy and robustness. Furthermore, it not only generates student innovation ability scores and classifications but also combines feature importance ranking and interpretability model analysis to reveal the contribution of each factor, effectively overcoming the black-box problem of traditional models and enhancing the scientific rigor and interpretability of the assessment results.

[0074] According to another aspect of the embodiments of this application, a storage medium is also provided. Optionally, in this embodiment, the storage medium can be used to execute the program code of any of the XGBoost innovation capability evaluation methods based on improved particle swarm optimization parameter tuning described above in the embodiments of this application.

[0075] Optionally, in this embodiment, the storage medium is configured to store program code for performing the following steps: S1, Collect multi-dimensional indicator data of the target object, and preprocess the collected multi-dimensional indicator data; S2, Principal component analysis was used to standardize the scale scores and behavioral numerical features and extract principal components; S3 uses XGBoost as the base learner and employs an improved particle swarm optimization algorithm to perform global search and optimization of the target hyperparameters. S4. Train the XGBoost model using the optimized hyperparameters, input new sample data to predict the innovation ability of the target object, and obtain the prediction result, which includes innovation ability score or level classification.

[0076] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated in this embodiment.

[0077] The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, as well as magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0078] According to another aspect of the embodiments of this application, an electronic device is also provided for implementing the above-described XGBoost innovation capability evaluation method based on improved particle swarm optimization parameter tuning. The electronic device may be a server, a terminal, or a combination thereof.

[0079] Figure 7 This is a schematic diagram of the structure of an optional electronic device according to an embodiment of this application, such as... Figure 7 As shown, it includes a processor 702, a communication interface 704, a memory 706, and a communication bus 708. The processor 702, communication interface 704, and memory 706 communicate with each other via the communication bus 708. Memory 706 is used to store computer programs; When processor 702 executes a computer program stored in memory 706, it performs the following steps: S1, Collect multi-dimensional indicator data of the target object, and preprocess the collected multi-dimensional indicator data; S2, Principal component analysis was used to standardize the scale scores and behavioral numerical features and extract principal components; S3 uses XGBoost as the base learner and employs an improved particle swarm optimization algorithm to perform global search and optimization of the target hyperparameters. S4. Train the XGBoost model using the optimized hyperparameters, input new sample data to predict the innovation ability of the target object, and obtain the prediction result, which includes innovation ability score or level classification.

[0080] Optionally, the communication bus can be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 The symbol is represented by a single thick line, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned electronic device and other devices.

[0081] The memory may include RAM, or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0082] The processors mentioned above can be general-purpose processors, including but not limited to: CPU (Central Processing Unit), NP (Network Processor), etc.; they can also be DSP (Digital Signal Processor), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0083] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0084] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0085] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0086] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0087] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of embodiments of this disclosure upon considering the specification and practicing the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

[0088] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0089] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for evaluating the innovation capability of XGBoost based on improved particle swarm optimization parameter tuning, characterized in that, include: Collect multidimensional indicator data of the target object, and preprocess the collected multidimensional indicator data; Principal component analysis was used to standardize the scale scores and behavioral numerical features and extract principal components. Using XGBoost as the base learner, an improved particle swarm optimization algorithm is used to perform global search and optimization of the target hyperparameters; The XGBoost model is trained using optimized hyperparameters, and new sample data is input to predict the innovation ability of the target object. The prediction results include innovation ability scores or level classifications.

2. The XGBoost innovation capability evaluation method based on improved particle swarm optimization parameter tuning as described in claim 1, characterized in that, The standardization and principal component extraction of scale scores and behavioral numerical features using principal component analysis include: Construct the correlation coefficient matrix and calculate the eigenvalues ​​and eigenvectors; Draw a scree plot and determine the number of principal components to be extracted based on the inflection point of eigenvalue descent. Calculate the cumulative explained variance, and determine the number of principal components using the cumulative explained variance and scree plot; Generate principal component loading matrices and interpret the semantics of each principal component by combining the variable loading values.

3. The XGBoost innovation capability evaluation method based on improved particle swarm optimization parameter tuning as described in claim 1, characterized in that, The target hyperparameters include: the learning rate, number of trees, maximum depth, row / column sampling ratio, minimum sample weight of leaf nodes, and regularization coefficient of the XGBoost model, all of which are automatically optimized within a preset range using the IPSO algorithm.

4. The XGBoost innovation capability evaluation method based on improved particle swarm optimization parameter tuning as described in claim 1, characterized in that, The method further includes: The prediction results are interpreted using SHAP values; Calculate the contribution of each feature to the prediction results and generate a ranking and visualization report of the factors influencing innovation capability.

5. The XGBoost innovation capability evaluation method based on improved particle swarm optimization parameter tuning as described in claim 1, characterized in that, The method of using XGBoost as the base learner and employing an improved particle swarm optimization algorithm to perform a global search and optimization of the target hyperparameters includes: The objective function of XGBoost is designed by progressively stacking multiple weak learners to form a strong learner. A second-order gradient approximation is performed, and the optimal split is determined by calculating the split gain.

6. The XGBoost innovation capability evaluation method based on improved particle swarm optimization parameter tuning as described in claim 1, characterized in that, The method of using XGBoost as the base learner and employing an improved particle swarm optimization algorithm to perform a global search and optimization of the target hyperparameters includes: An adaptive strategy based on cosine curves is used to adjust the inertia weights; An adaptive learning factor adjustment mechanism is adopted to make the group learning factor and individual learning factor change dynamically with the iteration process.

7. The XGBoost innovation capability evaluation method based on improved particle swarm optimization parameter tuning as described in claim 1, characterized in that, Before obtaining the prediction result by predicting the innovation ability of the target object using the new input sample data, the method further includes: Define the particle representation and fitness function, where each particle represents a candidate solution vector containing the target hyperparameters of XGBoost; An improved particle swarm optimization algorithm is used for iterative search, wherein the improved particle swarm optimization algorithm includes adaptive inertia weights and dynamic learning factors; When the iteration converges, the optimal particle solution vector is obtained, and the optimal parameter set and the optimal feature set are decoded.

8. An XGBoost innovation capability evaluation system based on improved particle swarm optimization parameter tuning, executing the XGBoost innovation capability evaluation method based on improved particle swarm optimization parameter tuning as described in any one of claims 1-7, characterized in that, include: The system includes a data acquisition module, a data preprocessing module, a model prediction module, and a results visualization module.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program, when executed, performs the method of any one of claims 1 to 7.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the method of any one of claims 1 to 7 through the computer program.

Images