Child myopia prevention and control prediction system and method based on data mining

Abstract

The invention discloses a child myopia prevention and control prediction system and method based on data mining, and relates to the technical field of health management and data mining, and the method comprises the steps: collecting dynamic modal data and static modal data of a child, and carrying out the preprocessing of the dynamic modal data and the static modal data; based on the two types of preprocessed modal data, dynamic modal features and static modal features of the child are extracted; on the basis of the dynamic modal features and the static modal features of the children, adopting a multi-modal fusion algorithm to generate personalized myopia risk features; constructing a myopia risk prediction model, and predicting the myopia probability of the child based on the personalized myopia risk features; and generating a personalized prevention and control strategy according to the myopia probability of the child. The deep neural network is adopted to construct the prediction model, the features are converted into the specific myopia probability, high-precision personalized prediction is provided, and early warning and customized prevention and control strategies are supported.

Description

Technical Field

[0001] The present invention relates to the technical field of health management and data mining, and in particular to a child myopia prevention and control prediction system and method based on data mining. Background Art

[0002] With the rapid development of information technology and the widespread application of data mining technology, data-driven health management methods have gradually become a research hotspot. As a high-incidence public health problem worldwide, the incidence of childhood myopia has continued to rise in recent years. Especially under the influence of the popularization of electronic devices and the increase in academic burden, the trend of myopia at a younger age has become more and more obvious. At present, it has become an important research direction to solve this problem by combining data mining technology with children's biological characteristics and behavioral data to predict the risk of myopia and formulate prevention and control strategies. Existing research mainly focuses on the analysis and modeling of single-modality data (such as physiological indicators such as axial length and refractive power) or behavioral statistics (such as reading time, screen time, etc.). However, since the occurrence and development of childhood myopia is the result of the combined action of multiple factors, single-modality data processing methods often cannot accurately capture the complex interactive relationship between multi-dimensional features, thereby limiting the accuracy of myopia risk prediction models and the personalization of prevention and control programs.

[0003] Although traditional myopia prediction methods based on statistical analysis or machine learning have achieved the assessment of myopia risk to a certain extent, the existing technology still has significant shortcomings. First, the existing methods have weak capabilities in fusing and processing multimodal data, and cannot fully utilize the complementary information of dynamic modal data (such as blinking frequency, head posture changes) and static modal data (such as axial length, refractive power, etc.). Dynamic modal data can reflect the real-time changes in children's daily eye behavior, while static modal data reflects the long-term trend of children's physiological characteristics. When the two types of data are not effectively fused, the prediction accuracy of the model is low. Secondly, most of the current myopia risk prediction models use simple linear regression or traditional machine learning methods, which are difficult to capture the nonlinear relationship between multimodal features, resulting in inaccurate assessment results of myopia risk, which further affects the pertinence and effectiveness of prevention and control strategies. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a child myopia prevention and control prediction system and method based on data mining to solve the problems of insufficient multimodal data fusion and limited nonlinear feature capture capability in the prior art.

[0006] The solution is as follows: In a first aspect, the present invention provides a method for predicting myopia prevention and control in children based on data mining, comprising: collecting dynamic modal data and static modal data of children, and preprocessing the two modal data, the dynamic modal data including the children's blinking frequency, pupil diameter changes, gaze time distribution and head posture changes, and the static modal data including the children's axial length, refractive power and reading distance; extracting the dynamic modal features and static modal features of the children based on the preprocessed two modal data, calculating the dot product of each dynamic modal feature and static modal feature with a weight vector, assigning weights to each dynamic modal feature and static modal feature, using a multimodal fusion algorithm to perform feature interaction calculations, and generating personalized myopia risk features; constructing a myopia risk prediction model based on a deep neural network, predicting the probability of myopia in children based on the personalized myopia risk features, and generating personalized prevention and control strategies.

[0007] As a preferred solution of the child myopia prevention and control prediction method based on data mining described in the present invention, the two modal data are preprocessed, and the specific steps are as follows: removing outliers in the dynamic modal data through the box plot method; using Gaussian filtering to smooth the dynamic modal data; converting the static modal data through standardization to eliminate different dimensions; and performing data dimensionality reduction processing on the static modal data through principal component analysis.

[0008] As a preferred solution of the method for predicting the prevention and control of myopia in children based on data mining described in the present invention, the specific steps of extracting the dynamic modal features and static modal features of children based on the preprocessed two modal data are as follows: Perform time series analysis on blink frequency data to extract the average blink frequency characteristics of children in a unit time; For pupil diameter change data, Fourier transform was used to extract the frequency components and fluctuation amplitude characteristics of children; For the head posture change data, extract the stability characteristics of the children's head posture in unit time; Time series analysis was performed on the axial length data to extract the growth characteristics of the children's axial length; Count the change amplitude and direction of refraction and extract the characteristics of children's refraction change; According to the preprocessed reading distance, the variability characteristics of children's reading distance were extracted using statistical analysis methods.

[0009] As a preferred solution of the method for predicting the prevention and control of myopia in children based on data mining described in the present invention, the dynamic modal features are weighted and the expression is: ; in, It is The weight coefficient of the dynamic modal features, It is The value of the dynamic mode feature, is the total number of dynamic mode features, is the weight vector transpose of the dynamic modal feature, is the index variable of the dynamic modal feature during the normalization process, It is the first The value of a dynamic modal feature; The weight distribution of static modal features is expressed as: ; Based on dynamic modal features, It is The weight coefficient of the static modal characteristics, It is The value of the static modal characteristic, is the total number of static modal features, is the weight vector transpose of the static modal features, is the index variable of the static modal characteristics during normalization, It is the first The value of a static modal characteristic; Weight coefficient and the weight coefficients of the static modal characteristics , a multimodal fusion algorithm is used to perform feature interaction calculation, capture the nonlinear relationship between each dynamic modal feature and the static modal feature, and generate personalized myopia risk features.

[0010] As a preferred solution of the method for predicting the prevention and control of myopia in children based on data mining described in the present invention, the expression of the personalized myopia risk feature is: ; in, is a personalized myopia risk profile, It is Dynamic modal eigenvalues The eigenvalues ​​obtained after nonlinear transformation are It is Static modal eigenvalues The eigenvalues ​​obtained after nonlinear transformation are is a positive constant that adjusts the difference between dynamic and static modal characteristics. is the regularization factor, It is The original value of the modal feature, It is The mean of the modal features, is the index variable of all dynamic modal features and static modal features.

[0011] As a preferred embodiment of the method for predicting the prevention and control of myopia in children based on data mining described in the present invention, the myopia risk prediction model is constructed based on a deep neural network, and the probability of myopia in children is predicted based on personalized myopia risk characteristics. The specific steps are as follows: Using DNN as a framework for myopia risk prediction models; Based on the framework of the myopia risk prediction model, the input layer, the first hidden layer, the second hidden layer and the output layer are constructed; The first hidden layer receives the personalized myopia risk features of the input layer , according to the weight matrix and bias vector, the personalized myopia risk characteristics Perform linear transformation and activation processing through ReLU activation function to generate preliminary feature representation; The second hidden layer receives the activation processing of repeated linear transformation and ReLU activation function of the preliminary feature representation to generate a deep feature representation; The output layer receives and generates deep feature representations and performs activation processing through the Sigmoid activation function; The input layer, the first hidden layer, the second hidden layer, the output layer and the framework of the myopia risk prediction model are combined to construct a myopia risk prediction model; Based on the personalized myopia risk characteristics, the probability of myopia in children is predicted, and the expression is: ; in, is the probability of myopia in children, is the weight matrix of the first hidden layer, is the weight matrix of the second hidden layer, is the weight matrix of the output layer, is the bias vector of the first hidden layer, is the bias vector of the second hidden layer, is the bias vector of the output layer.

[0012] As a preferred solution of the method for predicting the prevention and control of myopia in children based on data mining described in the present invention, the specific steps of generating a personalized prevention and control strategy are as follows: Based on the statistical analysis of the probability distribution of low risk and high risk in historical myopia population data, a low risk threshold P1 and a high risk threshold P2 are defined; when When ≤P1, the child is considered to be at low risk of myopia and should maintain the current healthy eye habits; When P1< When the value is ≤P2, the child’s myopia risk is considered to be at a medium risk, and the child should increase outdoor activity time and use a standardized reading lighting environment; when When ≥P1, the child’s current risk of myopia is considered to be high, and the time spent on electronic screens should be limited, myopia interventions should be conducted, and regular vision tests should be performed.

[0013] In the second aspect, the present invention provides a child myopia prevention and control prediction system based on data mining, including a data acquisition module, a feature extraction module, a feature fusion module, a myopia probability prediction module and a prevention and control strategy generation module; the data acquisition module is used to collect the dynamic modal data and static modal data of the child, and preprocess the dynamic modal data and the static modal data; the feature extraction module is used to extract the dynamic modal features and static modal features of the child based on the preprocessed two modal data; the feature fusion module is used to generate personalized myopia risk features based on the dynamic modal features and static modal features of the child using a multimodal fusion algorithm; the myopia probability prediction module is used to construct a myopia risk prediction model, and predict the myopia probability of the child based on the personalized myopia risk features; the prevention and control strategy generation module is used to generate a personalized prevention and control strategy according to the myopia probability of the child.

[0014] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the child myopia prevention and control prediction method based on data mining as described in the first aspect of the present invention.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the method for predicting the prevention and control of myopia in children based on data mining as described in the first aspect of the present invention.

[0016] The beneficial effects of the present invention are as follows: by fusing dynamic and static modal features and using the attention mechanism to generate personalized myopia risk features, accurate risk assessment is achieved. Then, a deep neural network is used to build a myopia risk prediction model, which converts these features into specific myopia probabilities, provides high-precision personalized predictions, and supports early warning and customized prevention and control strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0018] Figure 1 This is a flow chart of the method for predicting myopia prevention and control in children based on data mining in Example 1.

[0019] Figure 2 This is a schematic diagram of the children's myopia prevention and control prediction system based on data mining in Example 1. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0021] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides a method for predicting the prevention and control of myopia in children based on data mining, comprising the following steps: S1: Collect the dynamic modal data and static modal data of the child, and pre-process the dynamic modal data and static modal data.

[0024] Dynamic modality data include children’s blinking frequency, pupil diameter changes, gaze time distribution, and head posture changes; It should be noted that the collection of dynamic modal data mainly relies on smart wearable devices and screen monitoring software. Sensors worn on children, such as eye trackers and head posture sensors, can monitor blinking frequency, pupil diameter changes, and head posture changes in real time. These devices can capture data points at a high frequency to ensure the accuracy and continuity of the data. At the same time, through special software installed on electronic devices, children's gaze time distribution and screen usage time can be recorded, so as to fully understand children's eye behavior patterns.

[0025] Static modality data included children's axial length, refraction, and reading distance.

[0026] It should be noted that static modality data are obtained through regular medical examinations and questionnaires. The data of axial length and refractive power are usually measured by professional ophthalmologists using precision instruments and recorded in the child's health record. Reading distance can be collected through questionnaires filled out by parents or teachers, or assisted by smart devices (such as smart watches or mobile phone applications) to ensure the consistency and objectivity of the data. This method combines professional measurement with daily observation, providing a solid foundation for assessing children's vision health.

[0027] The outliers in the dynamic modal data were removed by using the box plot method; Furthermore, when processing dynamic modal data, the box plot method is first applied to identify and remove outliers. This method is based on statistical principles and uses the interval (IQR) between the first quartile (Q1) and the third quartile (Q3) to define data points within the normal range. Any data points outside this range are considered outliers and excluded, ensuring that the data set for subsequent analysis is purer and more accurate.

[0028] Use Gaussian filtering to smooth the dynamic modal data; Furthermore, Gaussian filtering technology is used to smooth the dynamic modal data. Gaussian filtering is a common image processing and data analysis technology. By applying a window with Gaussian distribution weights to the data points, the new value of each data point becomes the result of the weighted average of its neighboring points. This method can effectively reduce noise interference while retaining the main features and trends of the original data.

[0029] Static modal data is converted through standardization to eliminate different dimensions; Furthermore, a normalization process is applied to eliminate the effects of different dimensions. This process converts all features to the same scale, usually by subtracting the mean and dividing by the standard deviation. The normalized static modal data ensures that the features are comparable in subsequent analysis, avoiding bias caused by differences in units or magnitudes, thereby improving the accuracy of data analysis.

[0030] The principal component analysis method is used to reduce the dimension of static modal data.

[0031] Furthermore, principal component analysis (PCA) is used to reduce the dimensionality of static modal data. PCA is a statistical method used to identify linear relationships between variables in a data set and transform these variables into a new set of orthogonal variables, namely principal components. By selecting the first few principal components that explain most of the variance, the data dimension can be significantly reduced while retaining important information, simplifying the complexity of the subsequent modeling process.

[0032] S2: Extract children's dynamic modal features and static modal features based on the preprocessed two modal data.

[0033] Perform time series analysis on blink frequency data to extract the average blink frequency characteristics of children in a unit time; Furthermore, the time series analysis method is applied when processing the preprocessed blink frequency data. By analyzing the blink frequency data at a series of time points and calculating the average value in each unit time, the average blink frequency characteristics of children in that time period are extracted. This method can accurately capture the regular changes in children's blinking behavior and provide a stable basis for subsequent analysis.

[0034] For pupil diameter change data, Fourier transform was used to extract the frequency components and fluctuation amplitude characteristics of children; Furthermore, the Fourier transform technique is used to analyze the pupil diameter change data. This process converts the pupil diameter change in the time domain into a frequency domain representation to identify the frequency components and fluctuation amplitude characteristics. In this way, the periodic and non-periodic patterns of pupil diameter changes over time can be revealed, providing important information about children's visual responses.

[0035] For the head posture change data, extract the stability characteristics of the children's head posture in unit time; Furthermore, stability analysis is performed on the preprocessed head posture change data. By evaluating the changes in head posture per unit time, the stability characteristics of the head position are determined. This step helps to understand whether the child's head remains relatively still during the eye use process, and whether there are frequent or large posture changes, which is crucial to understanding children's eye use habits.

[0036] Time series analysis was performed on the axial length data to extract the growth characteristics of the children's axial length; Furthermore, time series analysis is also used when processing the preprocessed axial length data. By analyzing the axial length measurement results at different time points, the growth trend characteristics of the axial length are extracted. This process aims to track the changes in axial length over time and provide a key basis for evaluating the development of myopia in children.

[0037] Count the change amplitude and direction of refraction and extract the characteristics of children's refraction change; Furthermore, statistical analysis is performed on the refractive data to determine the magnitude and direction of the change. By quantifying the degree of change in refractive power and its increase or decrease trend, the refractive power change characteristics are extracted. This method can effectively reflect the dynamic changes in children's vision status and help identify possible vision problems.

[0038] According to the preprocessed reading distance, the variability characteristics of children's reading distance were extracted using statistical analysis methods.

[0039] Furthermore, based on the preprocessed reading distance data, statistical analysis methods were applied. By calculating the distribution of reading distances, including statistics such as minimum, maximum, and standard deviation, the variability characteristics of reading distances were extracted. This step aims to evaluate the changes in the distance between children's eyes and books or other reading materials during reading, and provide a reference for optimizing the eye environment.

[0040] It should be noted that the dynamic modal characteristics include: average blink frequency characteristics, frequency components and fluctuation amplitude characteristics of pupil diameter, and stability characteristics of head posture; static modal characteristics include axial length growth characteristics, diopter change characteristics, and variability characteristics of reading distance; S3: Based on children’s dynamic modal features and static modal features, a multimodal fusion algorithm is used to generate personalized myopia risk features.

[0041] Based on the dynamic modal features and static modal features of children, the dot product of each dynamic modal feature and static modal feature with the weight vector is calculated through the attention mechanism, and weights are assigned to each dynamic modal feature and static modal feature; Furthermore, when processing the dynamic modal features and static modal features of children, the attention mechanism is used to calculate the dot product of each feature and the weight vector, so as to assign corresponding weights to each feature. This process first inputs each dynamic modal feature and static modal feature into the attention model, and determines the importance of each feature by calculating the dot product between the feature vector and the weight vector. Subsequently, based on these dot product results, the dynamic modal features and static modal features respectively obtain corresponding weight values, so that in subsequent analysis, more critical features can be given higher attention.

[0042] The weight distribution of dynamic modal features is expressed as: ; in, It is The weight coefficient of the dynamic modal features, It is The value of the dynamic mode feature, is the total number of dynamic mode features, is the weight vector transpose of the dynamic modal feature, is the index variable of the dynamic modal feature during the normalization process, It is the first The value of a dynamic modal feature; It should be noted that when assigning weights to dynamic modal features, the dot product between each dynamic modal feature and the weight vector is first calculated. Then, the softmax function is applied to convert these dot products into weight coefficients in the form of probability distribution. Specifically, for the first The weight coefficient of a dynamic modal feature is obtained by calculating the dot product of its value and the weight vector, and dividing the result by the sum of the dot product exponential transformation values ​​corresponding to all dynamic modal features after exponential transformation. This process ensures that the sum of the weight coefficients of all dynamic modal features is 1, achieving normalization.

[0043] For example, when considering multiple dynamic modal features such as blinking frequency and pupil diameter changes, each feature will obtain a weight coefficient reflecting its relative importance according to the above formula. In this way, those features that better reflect children's eye habits and visual behavior patterns will be given higher weights, thus occupying a more important position in subsequent analysis. This method not only ensures the comparability between different features, but also enhances the accuracy and rationality of personalized myopia risk assessment.

[0044] The weight distribution of static modal features is expressed as: ; in, It is The weight coefficient of the static modal characteristics, It is The value of the static modal characteristic, is the total number of static modal features, is the weight vector transpose of the static modal features, is the index variable of the static modal characteristics during normalization, It is the first The value of a static modal characteristic; It should be noted that when weighting static modal features, the dot product between each static modal feature and the weight vector is first calculated. Then, the softmax function is applied to convert these dot products into weight coefficients in the form of probability distribution. Specifically, for For each static modal feature, the weight coefficient of the feature is obtained by calculating the dot product of its value and the weight vector, transforming the result through an exponential function, and dividing it by the sum of the dot product exponential transformation values ​​corresponding to all static modal features. This process ensures that the sum of the weight coefficients of all static modal features is 1, thus achieving normalization.

[0045] For example, when considering static modal features such as axial length and refractive power, each feature obtains a weight coefficient reflecting its relative importance according to the above formula. This means that those features that better reflect the long-term trend of children's vision changes will be given higher weights. This method not only ensures the comparability between different static modal features, but also enhances the accuracy and rationality of personalized myopia risk assessment. In this way, the importance of static modal features can be quantified, thereby better supporting subsequent analysis and prediction work.

[0046] Weight coefficient based on dynamic modal characteristics and the weight coefficients of the static modal characteristics , a multimodal fusion algorithm is used to perform feature interaction calculation, capture the nonlinear relationship between each dynamic modal feature and the static modal feature, and generate personalized myopia risk features. The expression is: ; in, is a personalized myopia risk profile, It is Dynamic modal eigenvalues The eigenvalues ​​obtained after nonlinear transformation are It is Static modal eigenvalues The eigenvalues ​​obtained after nonlinear transformation are is a positive constant that adjusts the difference between dynamic and static modal characteristics. is the regularization factor, It is The original value of the modal feature, It is The mean of the modal features, is the index variable of all dynamic modal features and static modal features.

[0047] The first The “modal features” mentioned in the original values ​​of the modal features include dynamic modal features and static modal features.

[0048] It should be noted that in the process of generating personalized myopia risk features, a multimodal fusion algorithm is used to process the weighted eigenvalues ​​through nonlinear transformation to capture the nonlinear relationship between dynamic and static modal features. Next, all weighted and transformed eigenvalues ​​are combined, and the importance of different features is balanced through normalization. Finally, an exponential decay function is applied to adjust the differential effects between features to generate the final personalized myopia risk feature R. This process integrates multiple feature information and improves the accuracy and personalization level of risk assessment.

[0049] S4: Construct a myopia risk prediction model to predict the probability of myopia in children based on personalized myopia risk characteristics.

[0050] Using DNN (Deep Neural Networks) as the framework of myopia risk prediction model; It should be noted that DNN is used as the framework of the myopia risk prediction model because DNN can automatically learn and extract nonlinear relationships between complex features, thereby providing more accurate and personalized myopia risk prediction.

[0051] Based on the framework of the myopia risk prediction model, the input layer, the first hidden layer, the second hidden layer and the output layer are constructed; The first hidden layer receives the personalized myopia risk features of the input layer , according to the weight matrix and bias vector, the personalized myopia risk characteristics Perform linear transformation and activation processing through ReLU activation function to generate preliminary feature representation; Furthermore, the first hidden layer receives the personalized myopia risk feature R from the input layer. The process is as follows: The personalized myopia risk feature R is passed to the first hidden layer, where each eigenvalue is multiplied by a weight matrix and a bias vector is added. This linear transformation operation is designed to capture the linear relationship between the input feature and the output. Subsequently, the transformed result is processed by the ReLU activation function, which sets all negative values ​​to zero while the positive values ​​remain unchanged. This step not only enhances the model's ability to learn nonlinear relationships, but also ensures that only meaningful feature representations are retained and enhanced. For example, if a child's personalized myopia risk feature indicates that his or her axial length is growing faster, the importance of this feature will be more prominent after ReLU activation.

[0052] The second hidden layer receives the activation processing of repeated linear transformation and ReLU activation function of the preliminary feature representation to generate a deep feature representation; Furthermore, the second hidden layer receives the preliminary feature representations generated from the first hidden layer, which have been processed by linear transformation and ReLU activation. In the second hidden layer, the same linear transformation is applied again to the preliminary feature representation, that is, the weight matrix and bias vector are applied again for transformation. Then, the result of the linear transformation is processed again using the ReLU activation function. This repeated process further abstracts the features and generates a deeper feature representation. For example, assuming that the first hidden layer identifies that an abnormally high blinking frequency may be one of the risk factors for myopia, the second hidden layer may further analyze the association between this high-frequency blinking and other eye behaviors (such as looking at the screen for a long time), thereby more accurately reflecting the potential risk of myopia.

[0053] The output layer receives and generates deep feature representations and performs activation processing through the Sigmoid activation function; Furthermore, the output layer receives the deep feature representation generated by the second hidden layer and is finally processed by the Sigmoid activation function. The Sigmoid function maps any real value to between 0 and 1, so that the output can be directly interpreted as a probability value. In this process, the information in the deep feature representation is integrated to produce a value between 0 and 1, representing the probability of myopia in children. For example, if a child's deep feature representation shows that both his or her eye habits and physiological characteristics point to a higher risk of myopia, the Sigmoid activation function will output a value close to 1, indicating a higher probability of myopia.

[0054] The input layer, the first hidden layer, the second hidden layer, the output layer and the framework of the myopia risk prediction model are combined to construct a myopia risk prediction model; Furthermore, the construction of the myopia risk prediction model combines the input layer, the first hidden layer, the second hidden layer, and the output layer. Each layer is connected in sequence to form a complete network structure for processing personalized myopia risk features and predicting the probability of myopia in children. The input layer receives the original personalized myopia risk features, the first hidden layer and the second hidden layer are responsible for gradually refining and abstracting these features, and the output layer generates the final prediction results.

[0055] Based on the personalized myopia risk characteristics, the probability of myopia in children is predicted, and the expression is: ; in, is the probability of myopia in children, is the weight matrix of the first hidden layer, is the weight matrix of the second hidden layer, is the weight matrix of the output layer, is the bias vector of the first hidden layer, is the bias vector of the second hidden layer, is the bias vector of the output layer.

[0056] It should be noted that after the model is built, the personalized myopia risk features are input into the myopia risk prediction model. Inside the model, the features first undergo linear transformation and ReLU activation processing in the first hidden layer, then undergo a deeper level of abstraction processing in the second hidden layer, and finally are converted into a specific myopia probability P through the Sigmoid activation function in the output layer. The entire process integrates multi-level information processing to ensure the accuracy and reliability of the prediction results. For example, for a child with specific eye habits and physiological characteristics, the model can output a specific myopia probability value based on their personalized myopia risk features to help parents or doctors take appropriate preventive measures.

[0057] S5: Generate personalized prevention and control strategies based on the probability of myopia in children.

[0058] Based on the statistical analysis of the probability distribution of low risk and high risk in historical myopia population data, a low risk threshold P1 and a high risk threshold P2 are defined; when When ≤P1, the child is considered to be at low risk of myopia and should maintain the current healthy eye habits; It should be noted that when the myopia probability P is less than or equal to the low-risk threshold P1, for example, the calculated myopia probability of a child is 0.2, which is far below the set low-risk threshold. This means that the child's current myopia risk is at a low level. In this case, it is recommended to continue to maintain existing healthy eye habits, such as maintaining an appropriate reading distance, resting the eyes regularly, and maintaining good lighting conditions. Parents do not need to take additional measures, just continue to pay attention to their children's vision.

[0059] When P1< When the value is ≤P2, the child’s myopia risk is considered to be at a medium risk, and the child should increase outdoor activity time and use a standardized reading lighting environment; It should be noted that when the myopia probability P is greater than the low risk threshold P1 but less than or equal to the medium risk threshold P2, for example, the myopia probability of a child is 0.6, which is between low risk and high risk. This indicates that the current myopia risk of children is at a medium level. In view of this situation, it is recommended to increase outdoor activity time, ensure at least a certain amount of natural light exposure every day, and ensure the use of standardized reading lighting environment to reduce the burden on the eyes. Through these preventive measures, the possibility of myopia development can be effectively reduced.

[0060] when When ≥P1, the child’s current risk of myopia is considered to be high, and the time spent on electronic screens should be limited, myopia interventions should be conducted, and regular vision tests should be performed.

[0061] It should be noted that when the myopia probability P is greater than or equal to the high-risk threshold P1, for example, the myopia probability of a child is as high as 0.85, which exceeds the set high-risk threshold. This means that the child is currently at a high risk of myopia. In this case, immediate action should be taken to limit the use of electronic screens, avoid long-term close-up use of the eyes, and consider professional myopia intervention measures, such as wearing corrective glasses or receiving visual training. In addition, regular vision examinations should be arranged to monitor vision changes in a timely manner and adjust treatment plans.

[0062] The present embodiment also provides a child myopia prevention and control prediction system based on data mining, including: a data acquisition module, a feature extraction module, a feature fusion module, a myopia probability prediction module and a prevention and control strategy generation module; the data acquisition module is used to collect the dynamic modal data and static modal data of the child, and pre-process the dynamic modal data and the static modal data; the feature extraction module is used to extract the dynamic modal features and static modal features of the child based on the pre-processed two modal data; the feature fusion module is used to generate personalized myopia risk features based on the dynamic modal features and static modal features of the child using a multimodal fusion algorithm; the myopia probability prediction module is used to construct a myopia risk prediction model, and predict the myopia probability of the child based on the personalized myopia risk features; the prevention and control strategy generation module is used to generate a personalized prevention and control strategy according to the myopia probability of the child.

[0063] This embodiment also provides a computer device, which is suitable for the case of a child myopia prevention and control prediction method based on data mining, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the child myopia prevention and control prediction method based on data mining as proposed in the above embodiment.

[0064] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0065] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for predicting the prevention and control of myopia in children based on data mining as proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, disk or optical disk.

[0066] In summary, the present invention achieves accurate risk assessment by fusing dynamic and static modal features and using the attention mechanism to generate personalized myopia risk features. Then, a deep neural network (DNN) is used to build a myopia risk prediction model, which converts these features into specific myopia probabilities, provides high-precision personalized predictions, and supports early warning and customized prevention and control strategies.

[0067] Example 2, referring to Table 1, is the second example of the present invention. To further verify the technical solution of the present invention, experimental simulation data of a child myopia prevention and control prediction method based on data mining is provided.

[0068] In order to verify the effectiveness of the prediction method for myopia prevention and control in children based on data mining, an experimental study was designed, which covered 120 children aged between 6 and 12 years old. Each child was evaluated for myopia using the method of the present invention and the traditional method: the experimental group (using the method of the present invention) and the control group (using the traditional evaluation method of the prior art). All children wore smart wearable devices such as eye trackers and head posture sensors, and used electronic devices with screen monitoring software installed. At the same time, medical examinations were performed regularly to collect static modal data, including axial length, refraction, etc., and the reading distance of children was determined through questionnaires.

[0069] First, in this experiment, children's dynamic modal data (such as blinking frequency and pupil diameter changes) and static modal data (such as axial length and diopter) were collected through smart wearable devices and screen monitoring software. The box plot method was used to remove outliers, and the dynamic data was smoothed using Gaussian filtering; the static data was standardized and the principal component analysis was used to reduce the dimension.

[0070] Secondly, based on the preprocessed data, dynamic modal features and static modal features were extracted. For example, time series analysis was performed on the blink frequency to extract the average blink frequency; Fourier transform was applied to the pupil diameter change data to extract frequency components and fluctuation amplitudes; the stability of the head pose was evaluated; the growth trend of the eye axis length was analyzed; the changes in refractive power were statistically analyzed; and the variability of the reading distance was calculated.

[0071] Then, personalized myopia risk features were generated through a multimodal fusion algorithm. The attention mechanism was used to assign weights to each feature and perform a non-linear transformation to capture the non-linear relationship between dynamic and static features, and finally, personalized myopia risk feature R was generated.

[0072] Finally, a deep neural network (DNN) model was constructed to predict the myopia probability P. This model received the personalized myopia risk feature R, underwent linear transformation and ReLU activation processing through two hidden layers, and output the myopia probability through the Sigmoid function in the output layer. At the same time, the low-risk threshold P1 was defined as 0.30, and the high-risk threshold P2 was defined as 0.70.

[0073] The control group adopted the traditional myopia risk assessment method, which was limited to simple feature superposition and did not apply advanced data mining techniques and multimodal fusion algorithms for personalized risk assessment.

[0074] Specifically, it is shown in Table 1 below: Table 1 Comparison Table of Experimental Data for Children's Myopia Risk Assessment Test subjects Blink frequency (times / minute) Pupil diameter change (mm) Axial length growth (mm / year) Diopters The predicted probability of myopia in the experimental group P The predicted probability of myopia in the control group P Actual myopia status (1 = myopia, 0 = no myopia) Children A 26.5 3.80 0.27 -1.00 0.78 0.25 1 Children B 15.2 3.20 0.18 -0.50 0.22 0.75 0 Children C 24.5 3.70 0.26 -0.92 0.83 0.78 1 By analyzing the experimental data, it can be seen that the prediction results of the experimental group of the present invention are significantly better than those of the control group. In Child A, the experimental group predicted the myopia probability to be 0.78 (high risk, P > P2), identifying it as high-risk myopia, which is consistent with the actual situation (myopia, actual value is 1), while the control group predicted it to be 0.25 (low risk, P < P1), misassessing it as low risk and underestimating the risk. In Child B, the experimental group predicted the myopia probability to be 0.22 (low risk, P < P1), judging it as low risk, which is consistent with the actual situation (not myopic, actual value is 0), while the control group predicted it to be 0.75 (high risk, P > P2), overestimating the risk incorrectly and misreporting it as high risk. In Child C, both the experimental group and the control group predicted it to be high risk (P > P2), but the predicted value of the experimental group was 0.83, which is closer to the actual situation (myopia, actual value is 1), and is better than the predicted value of 0.78 of the control group. The present invention accurately captures children's myopia risk features through multimodal data fusion and deep learning models, showing higher prediction accuracy and reliability.

[0075] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for predicting the prevention and control of myopia in children based on data mining, characterized by: include: Collecting dynamic modal data and static modal data of the child, and preprocessing the two modal data, the dynamic modal data includes the child's blinking frequency, pupil diameter change, gaze time distribution and head posture change, and the static modal data includes the child's axial length, diopter and reading distance; Based on the preprocessed two-modal data, the dynamic modal features and static modal features of children are extracted, the dot product of each dynamic modal feature and static modal feature with the weight vector is calculated, weights are assigned to each dynamic modal feature and static modal feature, and a multimodal fusion algorithm is used to perform feature interaction calculation to generate personalized myopia risk features; A myopia risk prediction model is constructed based on deep neural networks. Based on personalized myopia risk characteristics, the probability of myopia in children is predicted and personalized prevention and control strategies are generated.

2. The method for predicting myopia prevention and control in children based on data mining according to claim 1, characterized in that: The two modal data are preprocessed, and the specific steps are as follows: The outliers in the dynamic modal data were removed by using the box plot method; Use Gaussian filtering to smooth the dynamic modal data; Static modal data is converted through standardization to eliminate different dimensions; The principal component analysis method is used to reduce the dimension of static modal data.

3. The method for predicting myopia prevention and control in children based on data mining according to claim 2, characterized in that: The specific steps of extracting the dynamic modal features and static modal features of children based on the preprocessed two modal data are as follows: Perform time series analysis on blink frequency data to extract the average blink frequency characteristics of children in a unit time; For pupil diameter change data, Fourier transform was used to extract the frequency components and fluctuation amplitude characteristics of children; For the head posture change data, extract the stability characteristics of the children's head posture in unit time; Time series analysis was performed on the axial length data to extract the growth characteristics of the children's axial length; Count the change amplitude and direction of refraction and extract the characteristics of children's refraction change; According to the preprocessed reading distance, the variability characteristics of children's reading distance were extracted using statistical analysis methods.

4. The method for predicting the prevention and control of myopia in children based on data mining according to claim 3, characterized in that: The weight distribution of dynamic modal features is expressed as: ； in, It is The weight coefficient of the dynamic modal features, It is The value of the dynamic mode feature, is the total number of dynamic mode features, is the weight vector transpose of the dynamic modal feature, is the index variable of the dynamic modal feature during the normalization process, It is the first The value of a dynamic modal feature; The weight distribution of static modal features is expressed as: ； in, It is The weight coefficient of the static modal characteristics, It is The value of the static modal characteristic, is the total number of static modal features, is the weight vector transpose of the static modal features, is the index variable of the static modal characteristics during normalization, It is the first The value of a static modal characteristic; Weight coefficient based on dynamic modal characteristics and the weight coefficients of the static modal characteristics , a multimodal fusion algorithm is used to perform feature interaction calculation, capture the nonlinear relationship between each dynamic modal feature and the static modal feature, and generate personalized myopia risk features.

5. The method for predicting the prevention and control of myopia in children based on data mining according to claim 4, characterized in that: The expression of the personalized myopia risk feature is: ； in, is a personalized myopia risk profile, It is Dynamic modal eigenvalues The eigenvalues ​​obtained after nonlinear transformation are It is Static modal eigenvalues The eigenvalues ​​obtained after nonlinear transformation are is a positive constant that adjusts the difference between dynamic and static modal characteristics. is the regularization factor, It is The original value of the modal feature, It is The mean of the modal features, is the index variable of all dynamic modal features and static modal features.

6. The method for predicting the prevention and control of myopia in children based on data mining according to claim 5, characterized in that: The myopia risk prediction model is constructed based on a deep neural network, and the myopia probability of children is predicted based on personalized myopia risk characteristics. The specific steps are as follows: Using DNN as a framework for myopia risk prediction models; Based on the framework of the myopia risk prediction model, the input layer, the first hidden layer, the second hidden layer and the output layer are constructed; The first hidden layer receives the personalized myopia risk features of the input layer , according to the weight matrix and bias vector, the personalized myopia risk characteristics Perform linear transformation and activation processing through ReLU activation function to generate preliminary feature representation; The second hidden layer receives the activation processing of repeated linear transformation and ReLU activation function of the preliminary feature representation to generate a deep feature representation; The output layer receives and generates deep feature representations and performs activation processing through the Sigmoid activation function; The input layer, the first hidden layer, the second hidden layer, the output layer and the framework of the myopia risk prediction model are combined to construct a myopia risk prediction model; Based on the personalized myopia risk characteristics, the probability of myopia in children is predicted, and the expression is: ； in, is the probability of myopia in children, is the weight matrix of the first hidden layer, is the weight matrix of the second hidden layer, is the weight matrix of the output layer, is the bias vector of the first hidden layer, is the bias vector of the second hidden layer, is the bias vector of the output layer.

7. The method for predicting the prevention and control of myopia in children based on data mining according to claim 6, characterized in that: The specific steps of generating a personalized prevention and control strategy are as follows: Based on the statistical analysis of the probability distribution of low risk and high risk in historical myopia population data, a low risk threshold P1 and a high risk threshold P2 are defined; when When ≤P1, the child is considered to be at low risk of myopia and should maintain the current healthy eye habits; When P1< When the value is ≤P2, the child’s myopia risk is considered to be at a medium risk, and the child should increase outdoor activity time and use a standardized reading lighting environment; when When ≥P1, the child’s current risk of myopia is considered to be high, and the time spent on electronic screens should be limited, myopia interventions should be conducted, and regular vision tests should be performed.

8. A child myopia prevention and control prediction system based on data mining, based on the child myopia prevention and control prediction method based on data mining according to any one of claims 1 to 7, characterized in that: It includes data collection module, feature extraction module, feature fusion module, myopia probability prediction module and prevention and control strategy generation module; A data acquisition module, used for collecting dynamic modal data and static modal data of children, and preprocessing the dynamic modal data and static modal data; A feature extraction module, used for extracting dynamic modal features and static modal features of children based on the preprocessed two modal data; A feature fusion module is used to generate personalized myopia risk features based on the dynamic modal features and static modal features of children using a multimodal fusion algorithm; Myopia probability prediction module, which is used to build a myopia risk prediction model and predict the myopia probability of children based on personalized myopia risk characteristics; The prevention and control strategy generation module is used to generate personalized prevention and control strategies based on the probability of myopia in children.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the child myopia prevention and control prediction method based on data mining as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the child myopia prevention and control prediction method based on data mining as described in any one of claims 1 to 7 are implemented.

Images