Children mental health state evaluation system based on machine learning

Abstract

The invention discloses a children mental health state evaluation system based on machine learning, which belongs to the technical field of data processing and comprises a data integration module, an evaluation model construction module, an evaluation model parameter search module and a children mental health state evaluation module. The method specifically comprises the following steps: generating a plurality of sub-training sets to train weak classifiers, designing an adjustment function, constructing a dynamic weak classifier weight updating mechanism and a data weight updating mechanism, performing weighted combination on the weak classifiers to construct a strong classifier, and performing iterative training for multiple times to obtain an evaluation model; according to the method, influence factors are designed, associated derivative positions are generated, individual positions are updated by utilizing a guiding relation, replacement positions are generated by combining the current individual position with random disturbance of a search space range of the current individual position, part of the individual positions are selected for replacement, optimal parameters of an evaluation model are obtained, and complex child psychological health data are effectively coped with. And the accuracy and reliability of children mental health assessment are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and specifically refers to a children's mental health status assessment system based on machine learning. Background Art

[0002] The assessment of children's mental health status uses machine learning algorithms and data analysis techniques to evaluate and monitor children's mental health, accurately identify the current mental health status of children, continuously monitor the psychological changes of children, and timely discover potential psychological problems, so as to provide data support for professionals and assist in formulating appropriate treatment or intervention plans. However, in the existing children's mental health status assessment systems, there are problems of unbalanced children's mental health data and interference by noise, which cannot effectively capture the diverse characteristics of children's mental health, and it is difficult to comprehensively and accurately reflect the mental health status of children, resulting in inaccurate and incomplete assessment results; in the existing children's mental health status assessment systems, the characteristics of children's mental health data are complex and diverse, and individual differences cannot be fully considered, resulting in the problem that it is difficult to accurately identify the true mental health status of different children. Summary of the Invention

[0003] In view of the above situation, to overcome the defects of the existing technology, the present invention provides a child mental health status assessment system based on machine learning. In the existing child mental health status assessment system, there are problems such as unbalanced child mental health data and interference by noise, which cannot effectively capture the diverse characteristics of child mental health, and it is difficult to comprehensively and accurately reflect the mental health status of children, resulting in inaccurate and incomplete assessment results. In this solution, multiple sub-training sets are generated to train weak classifiers, better capturing child mental health data of different categories and reducing the assessment bias caused by class imbalance; an adjustment function is designed to construct a dynamic weak classifier weight update mechanism and data weight update mechanism to timely reflect the learning status of the model and better adapt to the changes in child mental health data; the weak classifiers are weighted and combined to construct a strong classifier, and the evaluation model is obtained through multiple iterative trainings, combining the characteristics of different weak classifiers, so that the final evaluation model can more comprehensively and accurately reflect the complexity of the child mental health status; in view of the problem that in the existing child mental health status assessment system, the characteristics of child mental health data are complex and diverse, and individual differences cannot be fully considered, resulting in difficulty in accurately identifying the true mental health status of different children, this solution designs influence factors based on the global optimal position and random position, generates associated derivative positions, and updates the individual positions using the guiding relationship, which can effectively adapt to the differences in child mental health data and avoid assessment errors caused by excessive differences; the probability of an individual position being replaced is calculated based on the fitness value, and a replacement position is generated by combining the current individual position with a random perturbation within its search space range, and some individual positions are selected for replacement to obtain the optimal parameters of the evaluation model, effectively dealing with the complex characteristics of child mental health data and improving the accuracy and reliability of child mental health assessment.

[0004] The child mental health status assessment system based on machine learning provided by the present invention includes a data integration module, an evaluation model construction module, an evaluation model parameter search module, and a child mental health status evaluation module;

[0005] The data integration module collects and processes historical child mental health data;

[0006] The evaluation model construction module trains weak classifiers by generating multiple sub-training sets, designs an adjustment function, constructs a dynamic weak classifier weight update mechanism and a data weight update mechanism, combines the weak classifiers with weights to construct a strong classifier, and obtains the evaluation model through multiple iterative trainings;

[0007] The evaluation model parameter search module designs influence factors, generates associated derivative positions, updates the individual positions using the guiding relationship, generates a replacement position by combining the current individual position with a random perturbation within its search space range, and selects some individual positions for replacement to obtain the optimal parameters of the evaluation model;

[0008] The child mental health status assessment module inputs real-time child mental health data into an assessment model constructed based on optimal parameters for classification, outputs classification labels, and obtains the current mental health status of the child.

[0009] Further, the data integration module collects and processes historical child mental health data;

[0010] The collection of historical child mental health data is to collect questionnaire survey data, sentiment analysis data, behavior data, physiological data, and mental health status;

[0011] The processing of historical child mental health data is to perform missing value filling, outlier removal, data normalization, and data encoding on the collected historical child mental health data, and construct a training data set and a test data set based on the historical child mental health data.

[0012] Further, the assessment model construction module constructs an assessment model based on machine learning. The assessment model is provided with a sub-training set generation unit, an iterative training unit, and a model determination unit, and specifically includes the following:

[0013] Sub-training set generation unit; perform random sampling with replacement on the training data set to generate Q sub-training sets, each sub-training set contains n data, and for each data in each sub-training set, the initial weight of the data is assigned as ;

[0014] Iterative training unit; assign the initial weight of each weak classifier as 1. Each iteration is to train Q weak classifiers, and combine the Q weak classifiers with weights to obtain a strong classifier. A total of T iterative trainings are performed. Each iterative training includes the following:

[0015] Train weak classifiers; for each sub-training set, based on the weight distribution of the current data, train weak classifiers based on support vector machines to obtain Q weak classifiers, and calculate the classification results and error rates of each weak classifier on the corresponding sub-training set;

[0016] Design an adjustment function ; According to the error rate of the current weak classifier, the weights of the weak classifiers in the previous iteration, and the scaling factor, obtain the adjustment function , calculate the adjustment amount for updating the weights; the formula used is as follows:

[0017] ;

[0018] In the formula, is the adjustment amount for updating its weight, is the weight of and They are the q-th weak classifiers during the t-th and (t - 1)-th iterative trainings respectively, and R q is the sub-training set corresponding to the q-th weak classifier, γ is the scaling factor, δ is the normalization constant, is the error rate of;

[0019] Update the weights of the weak classifiers; Based on the adjustment amount and the partial derivative of the adjustment amount with respect to the previous weights, construct a dynamic weak classifier weight update mechanism;

[0020] Weak classifier integration; Combine Q weak classifiers with weights to obtain a strong classifier;

[0021] Update the data weights; Combine the scaling factor with the classification results of the weak classifier and the strong classifier, and construct a dynamic data weight update mechanism considering multiple factors; The formula used is as follows:

[0022] ;

[0023] In the formula, and are the weights during the (t + 1)-th and t-th iterative trainings respectively of, is the i-th data in R q , is the classification label of to is H t the classification label of to is the weight of, is the normalization factor;

[0024] Model determination unit; Use the strong classifier H T obtained from the T-th training as the evaluation model.

[0025] Furthermore, the evaluation model parameter search module performs optimal parameter search on the penalty parameter and kernel function parameter of the support vector machine in the evaluation model based on the swarm intelligence algorithm, and is provided with a swarm initialization unit, a swarm update unit, a replacement unit and a model parameter determination unit, specifically including the following:

[0026] Swarm initialization unit; Establish a parameter search space based on the penalty parameter and kernel function parameter of the support vector machine in the evaluation model, and preset the swarm size A, fitness threshold z th and the maximum search times u max, randomly initialize the positions of A individuals within the parameter search space to form a population. Each individual position represents a set of model parameters. Use the exponential loss function of the evaluation model based on the model parameters for the test data set as the fitness value corresponding to the individual position;

[0027] Population update unit; generate associated derivative positions by designing influence factors and update the individual positions using the guiding relationship, thereby achieving the update of the entire population, including the following:

[0028] Design influence factor; design the influence factor based on the global optimal position and the random position; the formula used is as follows:

[0029] ;

[0030] ;

[0031] In the formula, is the position of the a-th individual at the u-th search, is 's influence factor, D sto is the random position, is the global optimal position at the u-th search. The global optimal position is the individual position with the smallest fitness value. rand is a random number generated with a uniform distribution within the range (0, 1), cos(·) is the cosine function, is from to 's included angle, is the L2 norm;

[0032] Generate associated derivative positions; generate an associated derivative position for each individual position according to the influence factor;

[0033] Update individual positions; use the mean of the current individual position and its associated derivative position as the guiding relationship to guide the update of the individual position;

[0034] Replacement unit; calculate the probability of each individual position being replaced based on the fitness value, sort all the individual positions in the population from high to low according to the probability, generate replacement positions by combining the current individual position with the random perturbation within its search space range, and select the top 10% of the individual positions for replacement according to the sorting result;

[0035] Model parameter determination unit; update the fitness value of the population. When the fitness value of the global optimal position is less than the fitness threshold z th , end the search, use the model parameters represented by the global optimal position as the optimal parameters, and construct an evaluation model based on the optimal parameters; otherwise, if the maximum search number u max, then transfer to the population initialization unit to re-initialize the individual positions; otherwise, increment the search count by 1 and transfer to the population update unit to continue the search.

[0036] Furthermore, the children's mental health status assessment module collects and processes real-time children's mental health data, inputs the processed real-time children's mental health data into an assessment model constructed based on optimal parameters for classification, outputs classification labels, and obtains the current mental health status of children.

[0037] The beneficial effects achieved by the present invention using the above solution are as follows:

[0038] (1) Aiming at the problems in the existing children's mental health status assessment system, such as unbalanced children's mental health data and being interfered by noise, which cannot effectively capture the diverse characteristics of children's mental health, making it difficult to comprehensively and accurately reflect the mental health status of children, resulting in inaccurate and incomplete assessment results. In this solution, multiple sub-training sets are generated to train the support vector machine to obtain weak classifiers, which can better capture the mental health data of different categories of children, effectively improve the recognition ability of the minority class, and reduce the assessment bias caused by class imbalance; according to the error rate of the current weak classifier, the weight of the previous iteration of the weak classifier, and the scaling factor, an adjustment function is designed to construct a dynamic weak classifier weight update mechanism and a data weight update mechanism. The dynamic adjustment mechanism can timely reflect the learning state of the model, better adapt to the changes in children's mental health data, and improve the real-time performance and reliability of the assessment results; the weak classifiers are weighted and combined to construct a strong classifier, and the evaluation model is obtained through multiple iterative trainings. By combining the characteristics of different weak classifiers, the final evaluation model can more comprehensively and accurately reflect the complexity of children's mental health status, reduce misjudgments, and ensure the accuracy of the assessment results.

[0039] (2) Aiming at the problems in the existing children's mental health status assessment system, such as the complex and diverse characteristics of children's mental health data, which cannot fully consider individual differences, resulting in difficulty in accurately identifying the true mental health status of different children. In this solution, each individual position represents a set of model parameters. Influence factors are designed based on the global optimal position and random positions to generate associated derivative positions, and the individual positions are updated using the guiding relationship, which can effectively adapt to the differences in children's mental health data, avoid assessment errors caused by excessive differences, and optimize the adaptability of the assessment model; based on the fitness value, the probability of an individual position being replaced is calculated, and replacement positions are generated by combining the current individual position with a random perturbation within its search space range. Part of the individual positions are selected for replacement to obtain the optimal parameters of the assessment model, and the model parameters that can accurately describe the mental health status of individual children are found, effectively coping with the complex characteristics of children's mental health data, and improving the accuracy and reliability of children's mental health assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Schematic diagram of the child mental health status evaluation system based on machine learning provided by the present invention;

[0041] Figure 2 Schematic diagram of the evaluation model construction module;

[0042] Figure 3 Schematic diagram of the evaluation model parameter search module.

[0043] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, and do not constitute a limitation to the present invention. Detailed implementation manners

[0044] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0045] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation to the present invention.

[0046] Example 1. Refer to Figure 1 , the child mental health status evaluation system based on machine learning provided by the present invention includes a data integration module, an evaluation model construction module, an evaluation model parameter search module, and a child mental health status evaluation module;

[0047] The data integration module collects and processes historical child mental health data, and sends the data to the evaluation model construction;

[0048] The evaluation model construction module receives the data sent by the data integration module, trains weak classifiers by generating multiple sub-training sets, designs an adjustment function, constructs a dynamic weak classifier weight update mechanism and a data weight update mechanism, combines the weak classifiers with weights to construct a strong classifier, and obtains an evaluation model through multiple iterative trainings, and sends the data to the evaluation model parameter search module;

[0049] The evaluation model parameter search module receives the data sent by the evaluation model construction module, designs influence factors, generates associated derivative positions, and updates the individual positions using the guiding relationship. It generates replacement positions by combining the current individual positions with random perturbations within their search space ranges, selects some individual positions for replacement to obtain the optimal parameters of the evaluation model, and sends the data to the child mental health status evaluation module;

[0050] The child mental health status evaluation module receives the data sent by the evaluation model parameter search module, inputs the real-time child mental health data into the evaluation model constructed based on the optimal parameters for classification, outputs classification labels, and obtains the current mental health status of the child.

[0051] Embodiment 2, refer to Figure 1 , this embodiment is based on the above embodiment. In the data integration module, historical child mental health data is collected and processed;

[0052] The collection of historical child mental health data is to collect questionnaire survey data, sentiment analysis data, behavior data, physiological data, and mental health status;

[0053] The questionnaire survey data includes the scores of the Child Behavior Checklist (CBCL) and the Children's Depression Inventory (CDI);

[0054] The sentiment analysis data includes voice sentiment analysis data and facial expression analysis data during the consultation process of children. The voice sentiment analysis data includes voice sentiment categories, tones, and speech rates, and the facial expression analysis data includes facial expression categories, the frequency of expression appearance, and the duration of expression;

[0055] The behavior data includes the average daily game behavior data and social behavior data of children in daily life. The game behavior data includes game duration, cooperation frequency, and competition behavior frequency, and the social behavior data includes social interaction frequency and interaction duration;

[0056] The physiological data includes sleep duration, sleep onset time, deep sleep duration, awakening times, heart rate, blood pressure, and skin conductance response;

[0057] The mental health status includes healthy, mild problems, moderate problems, and severe problems, and the mental health status is used as a data label;

[0058] The processed historical children's mental health data is to fill in missing values, remove outliers, normalize the data, and encode the data for the collected historical children's mental health data. Missing value filling uses the K-nearest neighbor imputation method to fill in missing values. Outlier removal uses the box plot method to identify and remove outliers. Data normalization uses the maximum-minimum scaling method to unify the data into the same range. Data encoding uses One-Hot encoding to convert data labels into numerical data, and a training dataset and a test dataset are constructed based on the historical children's mental health data.

[0059] Embodiment 3, refer to Figure 1 and Figure 2 , based on the above embodiment, in the evaluation model construction module, an evaluation model is constructed based on machine learning. The evaluation model is provided with a generated sub-training set unit, an iterative training unit, and a model determination unit, which specifically includes the following content:

[0060] Generated sub-training set unit; perform random sampling with replacement on the training dataset to generate Q sub-training sets, each sub-training set contains n data, and for each data in each sub-training set, the initial weight of the data is assigned as ; effectively reduces the evaluation bias caused by class imbalance and improves the accuracy of the overall evaluation result;

[0061] Iterative training unit; assign the initial weight of each weak classifier as 1. Each iteration is to train Q weak classifiers, and the Q weak classifiers are weighted and combined to obtain a strong classifier. A total of T iterative trainings are performed to flexibly adapt to the dynamic changes of children's mental health data and ensure the timeliness and reliability of the evaluation results; each iterative training includes the following content:

[0062] Train weak classifiers; for each sub-training set, based on the weight distribution of the current data, train weak classifiers based on support vector machines to obtain Q weak classifiers, and calculate the classification results and error rates of each weak classifier on the corresponding sub-training set; the formula for calculating the error rate is as follows:

[0063] ;

[0064] In the formula, is 's error rate, is the q-th weak classifier in the t-th iterative training, is the q-th weak classifier's corresponding sub-training set R q 's i-th data, is in the t-th iterative training 's weight, y i is 's true label, is The classification label for is an indicator function; when it is , it is 1, otherwise it is 0;

[0065] Design an adjustment function ; based on the error rate of the current weak classifier, the weight of the weak classifier in the previous iteration, and the scaling factor, obtain the adjustment function , and calculate the adjustment amount for updating the weight; the formula used is as follows:

[0066] ;

[0067] In the formula, is the adjustment amount for updating its weight, is the weight of and are the q-th weak classifiers in the t-th and (t - 1)-th iteration trainings respectively, R q is the sub-training set corresponding to the q-th weak classifier, γ is the scaling factor, γ > 0, δ is the normalization constant, , is the error rate of

[0068] Update the weak classifier weight; based on the adjustment amount and the partial derivative of the adjustment amount with respect to the previous weight, construct a dynamic weak classifier weight update mechanism; the formula used is as follows:

[0069] ;

[0070] In the formula, is the weight of is the partial derivative of with respect to ;

[0071] Weak classifier ensemble; combine Q weak classifiers with weights to obtain a strong classifier; the formula used is as follows:

[0072] ;

[0073] In the formula, H t is the strong classifier obtained from the t-th iteration training;

[0074] Update the data weight; combine the scaling factor with the classification results of the weak classifier and the strong classifier, and construct a dynamic data weight update mechanism by considering multiple factors; the formula used is as follows:

[0075] ;

[0076] Wherein, and are the weights during the (t + 1)-th and t-th iterative trainings respectively, the weight, is the i-th data in R q , is for the classification label, is t for the classification label, is the weight, is the normalization factor, , H t-1 is the strong classifier during the (t - 1)-th iterative training, is t-1 for the classification label, is the expectation for all data in R q ;

[0077] Model determination unit; The strong classifier H T obtained from the T-th training is used as the evaluation model, so that the final evaluation can reflect the diversity and complexity of children's mental health status.

[0078] By performing the above operations, aiming at the problems in the existing children's mental health status evaluation system that there are imbalances in children's mental health data and noise interference, it is impossible to effectively capture the diverse characteristics of children's mental health, it is difficult to comprehensively and accurately reflect the mental health status of children, resulting in inaccurate and incomplete evaluation results. This solution trains weak classifiers by generating multiple sub-training sets, better captures children's mental health data of different categories, effectively improves the recognition ability for minority classes, and reduces the evaluation bias caused by class imbalance; According to the error rate of the current weak classifier, the weight of the previous iterative weak classifier, and the scaling factor, a tuning function is designed to construct a dynamic weak classifier weight update mechanism and a data weight update mechanism. The dynamic adjustment mechanism can timely reflect the learning state of the model, better adapt to the changes in children's mental health data, and improve the timeliness and reliability of the evaluation results; The weak classifiers are weighted and combined to construct a strong classifier, and the evaluation model is obtained through multiple iterative trainings. By combining the characteristics of different weak classifiers, the final evaluation model can more comprehensively and accurately reflect the complexity of children's mental health status, reduce misjudgments, and ensure the accuracy of the evaluation results.

[0079] Example 4, refer to Figure 1 and Figure 3, based on the above embodiment, in the evaluation model parameter search module, the optimal parameter search for the penalty parameter and kernel function parameter of the support vector machine in the evaluation model is performed based on the swarm intelligence algorithm. There are a swarm initialization unit, a swarm update unit, a replacement unit, and a model parameter determination unit, which specifically include the following:

[0080] Swarm initialization unit; based on the penalty parameter and kernel function parameter of the support vector machine in the evaluation model, establish a parameter search space, and preset the swarm size A, fitness threshold z th and the maximum number of searches u max , randomly initialize the positions of A individuals within the parameter search space to form a swarm, use each individual position to represent a set of model parameters, and take the exponential loss function of the evaluation model established based on the model parameters for the test data set as the fitness value corresponding to the individual position;

[0081] Swarm update unit; generate associated derivative positions by designing an influence factor and update the individual positions using the guiding relationship, thereby realizing the update of the entire swarm, enabling the model to continuously adjust to adapt to the mental health characteristics of different children, including the following:

[0082] Design influence factor; design the influence factor based on the global optimal position and the random position; the formula used is as follows:

[0083] ;

[0084] ;

[0085] In the formula, is the position of the a-th individual at the u-th search, is 's influence factor, D sto is the random position, is the global optimal position at the u-th search, and the global optimal position is the individual position with the smallest fitness value. rand is a random number generated uniformly within the range of (0, 1), cos(·) is the cosine function, is the angle from to , is the L2 norm;

[0086] Generate associated derivative positions; generate an associated derivative position for each individual position according to the influence factor; the formula used is as follows:

[0087] ;

[0088] In the formula, is The associated derivative position, and randn generates random numbers from a normal distribution with a mean of 0 and a standard deviation of 1;

[0089] Update the individual position; use the mean of the current individual position and its associated derivative position as the guiding relationship to guide the update of the individual position; the formula used is as follows:

[0090] ;

[0091] In the formula, is the position of the a-th individual at the (u + 1)-th search;

[0092] Replacement unit; calculate the probability of each individual position being replaced based on the fitness value, sort all the individual positions in the population from high to low according to the probability, generate the replacement position by combining the current individual position with the random perturbation within its search space range, and select the top 10% of the individual positions for replacement according to the sorting result; the amplitude of the perturbation is adaptively adjusted according to the search progress, and as the number of iterations increases, the perturbation amplitude gradually decreases, which helps to better handle individual differences among children, thereby improving the robustness and adaptability of the evaluation model when dealing with complex mental health data; the formula used is as follows:

[0093] ;

[0094] ;

[0095] In the formula, is the replacement position of, where ub and lb are the upper and lower limits of the parameter search space respectively, is the probability of being replaced, is the fitness value of;

[0096] Model parameter determination unit; update the fitness value of the population. When the fitness value of the global optimal position is less than the fitness threshold z th , end the search, take the model parameters represented by the global optimal position as the optimal parameters, and construct an evaluation model based on the optimal parameters; otherwise, if the maximum search number u max is reached, then go to the population initialization unit to re-initialize the individual positions; otherwise, increment the search number by 1 and go to the population update unit to continue the search; this helps to automatically find the optimal parameter combination for the variable characteristics of children's mental health, thus ensuring that the evaluation model can accurately capture the mental health status of different children.

[0097] By performing the above operations, in view of the problems existing in the existing child mental health status evaluation system, where the characteristics of child mental health data are complex and diverse, and individual differences cannot be fully considered, resulting in difficulty in accurately identifying the true mental health status of different children, this solution uses each individual position to represent a set of model parameters, designs influence factors based on the global optimal position and random positions, generates associated derivative positions, and updates the individual positions using the guiding relationship, which can effectively adapt to the differences in child mental health data, avoid evaluation errors caused by excessive differences, and optimize the adaptability of the evaluation model; calculates the probability of an individual position being replaced based on the fitness value, generates a replacement position by combining the current individual position with a random perturbation within its search space range, selects some individual positions for replacement, obtains the optimal parameters of the evaluation model, finds the model parameters that can accurately describe the mental health status of a child individual, effectively cope with the complex characteristics of child mental health data, and improve the accuracy and reliability of child mental health evaluation.

[0098] Example 5. Refer to Figure 1 , based on the above example, in the child mental health status evaluation module, real-time child mental health data is collected and processed. The collection of real-time child mental health data is to collect questionnaire survey data, sentiment analysis data, behavior data, and physiological data. The processing of real-time child mental health data is to perform data normalization processing on the collected real-time child mental health data, input the processed real-time child mental health data into the evaluation model constructed based on the optimal parameters for classification, output classification labels, and obtain the current mental health status of the child.

[0099] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0100] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention.

[0101] The above description of the present invention and its implementation manners is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. In general, if those of ordinary skill in the art are inspired by it and design, without creative efforts, structural manners and embodiments similar to the technical solution without departing from the gist of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. A children's mental health status assessment system based on machine learning, characterized by: It includes data integration module, assessment model building module, assessment model parameter search module and children's mental health status assessment module; The data integration module collects and processes historical children's mental health data; The evaluation model construction module generates multiple sub-training sets to train weak classifiers, designs adjustment functions, constructs dynamic weak classifier weight update mechanisms and data weight update mechanisms, weights weak classifiers to construct strong classifiers, and iterates training multiple times to obtain an evaluation model; The evaluation model parameter search module designs influencing factors, generates associated derivative positions, and uses guided relationships to update individual positions, generates replacement positions by combining the current individual positions with random perturbations of their search space range, selects some individual positions for replacement, and obtains the optimal parameters of the evaluation model; The children's mental health status assessment module inputs real-time children's mental health data into an assessment model constructed based on optimal parameters for classification, outputs classification labels, and obtains the children's current mental health status.

2. The children's mental health status assessment system based on machine learning according to claim 1 is characterized by: The evaluation model construction module constructs an evaluation model based on machine learning. The evaluation model is provided with a sub-training set generation unit, an iterative training unit and a model determination unit, which specifically include the following contents: Generate sub-training set units; perform random sampling with replacement on the training data set to generate Q sub-training sets, each of which contains n data. For each data in each sub-training set, the initial weight of the data is assigned ; Iterative training unit: the initial weight of each weak classifier is assigned to 1. Each iteration trains Q weak classifiers and weightedly combines the Q weak classifiers to obtain a strong classifier. A total of T iterative trainings are performed. Each iterative training includes the following contents: Train weak classifiers; for each sub-training set, under the weight distribution of the current data, train the weak classifier based on the support vector machine to obtain Q weak classifiers, and calculate the classification result and error rate of each weak classifier on the corresponding sub-training set; Design adjustment function ; According to the error rate of the current weak classifier, the weight of the weak classifier of the previous iteration and the proportional factor, the adjustment function is obtained, and the adjustment amount used to update the weight is calculated; the formula used is as follows: ； In the formula, Is used to update The amount of adjustment of its weight, yes The weight of and are the qth weak classifiers in the tth and t-1th iteration training, respectively, R q is the sub-training set corresponding to the qth weak classifier, γ is the scaling factor, δ is the standardization constant, yes The error rate; Update the weights of weak classifiers; construct a dynamic weak classifier weight update mechanism based on the adjustment amount and the partial derivative of the adjustment amount with respect to the previous weight; Weak classifier integration: Q weak classifiers are weighted together to obtain a strong classifier; Update data weights; Model determination unit: The strong classifier H obtained by the T-th training T as an evaluation model.

3. The children's mental health status assessment system based on machine learning according to claim 2 is characterized by: The updating of data weight is to combine the proportional factor with the classification results of the weak classifier and the strong classifier, and to build a dynamic data weight updating mechanism by integrating multiple factors; the formula used is as follows: ； In the formula, and They are the t+1th and tth iteration training respectively. The weight of YesR q The i-th data in yes right The classification label of Yes H t right The classification label of is the normalization factor, yes The weight of .

4. The children's mental health status assessment system based on machine learning according to claim 1 is characterized by: The evaluation model parameter search module searches for optimal parameters of the penalty parameters and kernel function parameters of the support vector machine in the evaluation model based on the swarm intelligence algorithm, and is provided with a swarm initialization unit, a swarm update unit, a replacement unit and a model parameter determination unit, and specifically includes the following contents: Population initialization unit; establish parameter search space based on the penalty parameters and kernel function parameters of the support vector machine in the evaluation model, and pre-set the population size A and fitness threshold z th and the maximum number of searches u max , randomly initialize A individual positions in the parameter search space to form a group, use each individual position to represent a set of model parameters, and use the exponential loss function of the evaluation model established based on the model parameters for the test data set as the fitness value of the corresponding individual position; Group update unit: Generates associated derivative positions by designing influencing factors and updates individual positions using guiding relationships, thereby updating the entire group, including the following: Design impact factors; Design impact factors based on global optimal positions and random positions; Generate associated derived positions; generate an associated derived position for each individual position according to the impact factor; Update individual positions; use the mean of the current individual position and its associated derivative position as a guiding relationship to guide the individual position to complete the update; Replacement unit; Model parameter determination unit; update the fitness value of the group, when the fitness value of the global optimal position is less than the fitness threshold z th When , the search ends, the model parameters represented by the global optimal position are taken as the optimal parameters, and the evaluation model is constructed based on the optimal parameters; otherwise, if the maximum number of searches u is reached max , then go to the group initialization unit to re-initialize the individual position; otherwise, add 1 to the search times and go to the group update unit to continue searching.

5. The children's mental health status assessment system based on machine learning according to claim 4 is characterized in that: The design impact factor is based on the global optimal position and random position design impact factor; the formula used is as follows: ； ； In the formula, is the position of the ath individual in the uth search, yes The impact factor, D sto is a random position, is the global optimal position at the u-th search, rand is a uniformly distributed random number generated in the range (0, 1), cos(·) is the cosine function, Is arrive The angle of is the L2 norm.

6. The children's mental health status assessment system based on machine learning according to claim 4 is characterized in that: The replacement unit calculates the probability of each individual position being replaced based on the fitness value, sorts all individual positions in the group from high to low according to the probability, generates replacement positions by combining the current individual position with the random perturbation of its search space range, and selects the top 10% of individual positions for replacement according to the sorting results.

7. The children's mental health status assessment system based on machine learning according to claim 1 is characterized by: The data integration module is used to collect and process historical children's mental health data; The collecting of historical children's mental health data includes collecting questionnaire data, sentiment analysis data, behavior data, physiological data and mental health status; The processing of historical children's mental health data is to fill missing values, remove outliers, normalize data and encode data on the collected historical children's mental health data, and construct a training data set and a test data set based on the historical children's mental health data.

8. The children's mental health status assessment system based on machine learning according to claim 1 is characterized by: The children's mental health status assessment module collects and processes real-time children's mental health data, inputs the processed real-time children's mental health data into an assessment model constructed based on optimal parameters for classification, outputs classification labels, and obtains the child's current mental health status.

Images