4G sphygmomanometer and blood pressure artificial intelligence analysis system

By using a 4G AI blood pressure monitor to record and analyze users' blood pressure, pulse, mood, exercise, and medication information, personalized health reports and suggestions are generated, solving the problem that home blood pressure monitors cannot record and analyze data, and enabling personalized health management and optimization of medical resources.

CN119033344APending Publication Date: 2024-11-29LONGVON TECH (SHENZHEN) CO LTD
View PDF 0 Cites 5 Cited by

Patent Information

Application Number
CN202411142227.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-20
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

Existing home blood pressure monitors cannot record and analyze daily monitoring results or provide health advice, thus failing to meet users' personalized health management needs.

Method used

Design a 4G AI blood pressure monitor that uses a digital blood pressure detection module, a data communication module, and a display module, combined with large-scale AI analysis, to record and analyze the user's blood pressure, pulse, mood, exercise status, and medication status, generate health reports and suggestions, and support data sharing.

Benefits of technology

It enables personalized health management, prevents chronic diseases, optimizes medical resources, supports telemedicine, and improves users' health management level and the efficient use of medical resources.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119033344A_ABST
    Figure CN119033344A_ABST
Patent Text Reader

Abstract

The invention discloses a 4G sphygmomanometer and a blood pressure artificial intelligence analysis system, and relates to improvement of intelligent medical detection equipment, in particular to improvement of an artificial intelligence sphygmomanometer. The method comprises the following steps of 1, user data acquisition and input, 2, user data storage and management, 3, user data analysis and modeling, 4, user blood pressure data report generation and health suggestion report generation, 5, user interaction and feedback, and 6, model optimization and updating. Measurement results are recorded through a household digital blood pressure measuring instrument, pulse, user mood, motion states, medicine taking conditions and the like are synchronously recorded, daily blood pressure report, weekly blood pressure report, monthly blood pressure report, monthly blood pressure report, annual blood pressure report and the like are formed, and corresponding health reports, health suggestions, medicine taking advice and the like are given through artificial intelligence analysis of a large model. And the blood pressure information of relatives and friends can be published and shared through corresponding software, so that the blood pressure condition care of relatives and friends is realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the improvement of intelligent medical detection equipment, and in particular to the improvement of an artificial intelligence sphygmomanometer. Background Art

[0002] Blood pressure is vital to human health because it directly affects the normal function of the cardiovascular system. Blood pressure is an important indicator of cardiovascular health. Maintaining normal blood pressure is essential for preventing diseases, prolonging life and improving quality of life. Regularly monitoring and controlling blood pressure is an important measure to maintain overall health.

[0003] Blood pressure is also one of the important indicators of human health. The main indicators of blood pressure are: systolic pressure, diastolic pressure, pulse, pulse pressure difference, and mean arterial pressure. Currently, the commonly used blood pressure meters are mercury sphygmomanometers used in hospitals, clinics and other places. This type of equipment requires professional operation by medical personnel to measure, and cannot allow users to perform daily and home tests.

[0004] There are also some home blood pressure monitors available. These blood pressure monitors can measure blood pressure, but they are unable to record, analyze and give corresponding suggestions for daily monitoring results, and are unable to achieve more functions. Summary of the invention

[0005] The purpose of the present invention is to provide a 4G artificial intelligence sphygmomanometer in response to the defects and shortcomings of the prior art. It records the measurement results through a home digital blood pressure measuring instrument and simultaneously records the pulse, user mood, exercise status, medication status, etc., to form daily, weekly, monthly, quarterly, annual reports, etc. of blood pressure, and gives corresponding health reports, health advice, and medication and medical advice through artificial intelligence analysis of a large model. It can also publish and share blood pressure information of relatives and friends through corresponding software, so as to care for the blood pressure status of relatives and friends.

[0006] To achieve the above-mentioned purpose, the 4G sphygmomanometer in the present invention adopts the following technical scheme: it includes a digital blood pressure detection module 1, a data communication module 2, and a display module 3. The blood pressure detection information of the digital blood pressure detection module 1 is connected to the 4G network or the wifi network through the data communication module 1 to connect to the smart device 4, and the blood pressure detection information displays relevant information through the display module 3; the smart device 4 records and analyzes blood pressure information, customer information, user mood information, exercise information, and medication information through a large data model and artificial intelligence, and generates through artificial intelligence: health reports, diet adjustment suggestions, exercise suggestions, and work and rest regularity suggestions.

[0007] The 4G sphygmomanometer artificial intelligence blood pressure analysis system of the present invention adopts the following steps: 1. user data collection and entry, 2. user data storage and management, 3. user data analysis and modeling, 4. user blood pressure data report generation and health advice report, 5. user interaction and feedback, 6. model optimization and update;

[0008] The step 1 of collecting and entering user data includes: collecting the systolic pressure, diastolic pressure, pulse pressure difference and other data of the user's blood pressure in real time; recording the user's heart beats per minute and synchronizing it with the blood pressure data; manually inputting the user's mood information, and the user's mood can be set to several levels including: very happy, happy, normal, unhappy, and very unhappy; the user's exercise type, duration, intensity, and interval time before measuring blood pressure, the user's exercise information can be manually collected or automatically entered by smart devices such as bracelets; recording the user's medication use on the day, including the drug name, dosage, time, etc.

[0009] The user data storage and management in step 2 includes: designing special tables or data sets for different categories of data, such as blood pressure data tables, pulse data tables, mood data tables, etc.; cleaning and preprocessing the data, cleaning the collected data, removing abnormal values ​​and missing values, and standardizing the data format to ensure the consistency and accuracy of the data;

[0010] The data cleaning is as follows: 2.1. The processing of missing values ​​is as follows: 2.1.1. Delete missing values. If the missing value ratio is low, such as less than 5% of the data, delete these records directly to ensure the integrity of the data set; 2.1.2. Use interpolation to process missing values: For time series data, such as blood pressure and pulse, linear interpolation, spline interpolation and other methods can be used to fill in missing values.

[0011] The linear interpolation formula is:

[0012] Among them, x1 and x2 are the values ​​at time points t1 and t2 respectively, and t is the time point of the missing value;

[0013] 2.1.3. For non-time series data, use the mean or median of the same feature column to fill missing values ​​to reduce data bias;

[0014] 2.1.4 Use the K-nearest neighbor algorithm to fill missing values ​​based on the values ​​of similar data points, which is suitable for high-dimensional data;

[0015] 2.2. The method for handling outliers is as follows: 2.2.1 Statistical analysis: outliers are detected by box plot and Z score statistical method;

[0016] Z-score formula:

[0017] Where x is a data point, μ is the mean, and σ is the standard deviation. Generally, data points with a Z score greater than 3 or less than -3 can be considered outliers. 2.2.2. Delete outliers. For extreme outliers, delete them directly to avoid misleading the analysis results. 2.2.3. Replace outliers: Outliers can be replaced with adjacent values ​​or medians to smooth the data distribution. 2.2.4. Boundary processing: If the data has physical or logical boundaries (such as blood pressure data will not be lower than 30 mmHg or higher than 300 mmHg), the values ​​outside the boundaries will be adjusted to the boundary values.

[0018] 2.3. The method for handling duplicate data is as follows: 2.3.1. Identify duplicate records and check the duplicate records in the data set through primary keys or composite keys (such as timestamps, user IDs, etc.); 2.3.2. Delete duplicate records. If the duplicate records are exactly the same, delete only one and keep only one; if the data is slightly different, keep the newer or more complete data record.

[0019] The data preprocessing is specifically as follows:

[0020] 2.4. Data standardization;

[0021] 2.5. Discretize the data, including: 2.5.1. Equal-width binning, dividing continuous data into intervals of equal width, and the number of data points in each interval may be different; 2.5.2. Equal-frequency binning, dividing the data into intervals of equal frequency, and the number of data points in each interval is the same; 2.5.3. Custom binning, manually define the binning intervals according to business needs or data distribution, such as dividing blood pressure levels into three intervals: low pressure, normal pressure, and high pressure.

[0022] 2.6. Smooth the data.

[0023] 2.7. Feature engineering, including feature selection: selecting features that have a greater impact on the target variable based on their importance (e.g., through correlation analysis, information gain, etc.); feature construction: generating new features by combining and transforming existing features, such as constructing a health score from blood pressure and pulse data; feature encoding: encoding categorical features, such as converting mood data into numerical form, using one-hot encoding or label encoding;

[0024] 2.8. Data verification, including: data distribution check, by drawing histograms, scatter plots, etc. to check data distribution and verify the effect of data cleaning and preprocessing; model performance evaluation, using part of the cleaned data for model training and testing, and evaluating the impact of data preprocessing on model performance;

[0025] 2.9. Storing and updating data, including data after cleaning: storing the cleaned and preprocessed data in a database or data warehouse for subsequent analysis and modeling; regular updating and maintenance: regularly cleaning and preprocessing new data to ensure the continued stability of data quality. Data cleaning and preprocessing in step 2 are the basic steps for data analysis and modeling. They ensure the quality of the data and lay a solid foundation for subsequent analysis and model training. Through these steps, data noise and errors can be minimized and the accuracy and reliability of the model can be improved.

[0026] The specific steps of step 3 user data analysis and modeling are as follows:

[0027] 3.1. Trend analysis: weekly, monthly, quarterly and annual blood pressure analysis. Specifically, the moving average method is used to smooth short-term fluctuations and show long-term trends.

[0028] The moving average method formula is:

[0029] Among them, P t is the blood pressure value on day t, and n is the time window size;

[0030] 3.2. Data aggregation: aggregate daily blood pressure, pulse, mood, exercise, and medication data, and calculate daily health scores through weighted average;

[0031] The weighted average health score formula is:

[0032] HealthScore=w1×BloodPressureScore+w2×PulseScore+w3×

[0033] MoodScore+w4×ExerciseScore+w5×MedicationScore

[0034] Among them, w1, w2, ..., w5 are the weights of each indicator, which can be adjusted through data training.

[0035] 3.3. Anomaly detection, using statistical analysis and machine learning algorithms (such as outlier detection algorithms) to identify abnormal changes in blood pressure, such as sudden increases or decreases;

[0036] Outlier detection formula:

[0037] Detect outliers by standard deviation and mean: Threshold = μ ± 2σ

[0038] Where μ is the mean value of blood pressure and σ is the standard deviation;

[0039] 3.4. User health risk assessment: Based on long-term data, combined with the user's historical data and health records, the user's health risk is assessed through algorithms such as logistic regression and random forest, and future blood pressure trends are predicted;

[0040] The logistic regression model formula is as follows:

[0041]

[0042] Where Y is the health risk result, X1, X2, X3…, X n is the input feature.

[0043] The step 5 user interaction and feedback includes 5.1. Real-time monitoring: through mobile or web applications, users can view their blood pressure data and health reports in real time and get personalized suggestions; 5.2. User feedback mechanism: users can provide feedback on their physical condition or health information, and the model will adjust itself based on these feedbacks to improve the accuracy of the analysis.

[0044] The step 6 is the optimization and updating of the model; including 6.1. Continuous optimization of the machine learning model: as the amount of data increases, the model is continuously trained and optimized using new data to improve prediction accuracy; 6.2. Algorithm adjustment: based on user feedback and actual results, the parameters and weights in the algorithm are adjusted to adapt to individual differences and new health trends.

[0045] The continuous optimization and algorithm adjustment of the machine learning model in step 6.1 is an important step to improve the prediction accuracy and stability. The commonly used optimization methods and specific model information are as follows:

[0046] 6.1.1. Model selection and initial construction:

[0047] Model selection: In the initial stage, you may choose different models to try, such as Logistic Regression, Decision Tree, Random Forest, Support Vector Machine (SVM), Neural Network, etc.

[0048] Initial model training: Use the cleaned and preprocessed data set to perform initial training on the selected model and evaluate its basic performance (such as accuracy, precision, recall, etc.).

[0049] 6.1.2. Model performance evaluation and tuning are:

[0050] Cross-Validation: Use K-fold cross-validation (such as 5-fold or 10-fold) to evaluate the generalization ability of the model and reduce the risk of overfitting. Process: Divide the dataset into K subsets, use one of the subsets as the validation set in turn, and the remaining K-1 subsets as the training set, repeat K times, and finally take the average performance.

[0051] 6.1.3. Feature engineering optimization is:

[0052] Feature Selection: Select features that have a greater impact on the target variable based on feature importance (such as through Lasso regression, recursive feature elimination (RFE), etc.), and eliminate irrelevant or redundant features.

[0053] Feature construction and transformation: Use polynomial features, interactive features, logarithmic transformation and other methods to enhance the expressiveness of the model and capture nonlinear relationships.

[0054] Dimensionality Reduction: Use PCA (Principal Component Analysis) or LDA (Linear Discriminant Analysis) and other methods to reduce the dimensionality of high-dimensional data to reduce the complexity of the model and the risk of overfitting.

[0055] 6.1.4. Model integration and improvement:

[0056] Ensemble Learning:

[0057] Combine the prediction results of multiple models to improve the robustness and prediction performance of the model.

[0058] 6.1.5. Model monitoring and feedback are adjusted to:

[0059] Model Monitoring: Continuously monitor the model’s predictive performance, especially on new data, to see if the model is experiencing performance degradation or increased bias;

[0060] Feedback Adjustment: Dynamically adjust the model’s weights, hyperparameters, or data processing methods based on the results of model monitoring to ensure that the model continues to maintain high performance;

[0061] Online learning: For real-time streaming data, use online learning algorithms (such as SGD and OnlineSVM) to gradually update the model to keep the model adaptable to the latest data;

[0062] Update the model parameters using gradient descent:

[0063] θ t+1=θ t -η▽ θ L(θ t ;x t ,y t ).

[0064] 6.1.6. Algorithm adjustment and model update are as follows:

[0065] Retraining the model: When the data set expands significantly or the features change, it may be necessary to retrain or update the existing model; the specific retraining methods are: Incremental Learning: gradually update the weights based on the existing model; Periodic retraining: regularly retrain the model with the latest data set to adapt to new situations.

[0066] Algorithm upgrade: With the development of technology and the increase in data volume, more complex models or algorithms (such as deep learning and reinforcement learning) may be needed to improve predictive capabilities.

[0067] 6.1.7. Model validation and evaluation: Model validation: Use the validation set or test set to evaluate the final performance of the model. Common indicators include AUC, F1-score, precision, recall, etc. Model deployment and launch: Deploy the final optimized model to the production environment to provide real-time or batch prediction services.

[0068] The continuous optimization and algorithm adjustment of machine learning models is an iterative process, including model selection, feature engineering, ensemble learning, hyperparameter tuning, feedback adjustment, etc. Through these methods, the prediction performance of the model can be continuously improved to ensure the accuracy and reliability of the model in practical applications.

[0069] The working principle of the present invention is as follows: it uses a home-use digital blood pressure measuring instrument to record the measurement results and simultaneously record the pulse, user's mood, exercise status, medication status, etc., to form daily, weekly, monthly, quarterly, and annual blood pressure reports, etc., and uses artificial intelligence analysis of a large model to give corresponding health reports, health advice, and medication and medical advice, etc., and can use corresponding software to publish and share blood pressure information of relatives and friends, and use the software's own social functions to care for the blood pressure status of relatives and friends.

[0070] After adopting the above technical solution, the beneficial effects of the present invention are: it provides practical advantages in personalized health management, prevention of chronic diseases, optimization of medical resources, support for telemedicine, etc. These advantages not only improve the health management level of users, but also promote the efficient use of medical resources and promote the intelligent and personalized development of health management.

[0071] It can achieve personalized health management. Through long-term monitoring and analysis of multi-dimensional information such as the user's blood pressure, pulse, mood, exercise and medication, the model can generate personalized health reports, including diet, work and rest, exercise and medication recommendations. These recommendations can be adjusted according to the user's specific health status to improve management effectiveness; abnormal changes can be detected in a timely manner, and health management plans can be adjusted dynamically. For example, when the model detects that the user's blood pressure fluctuates abnormally, it may recommend increasing exercise, adjusting diet or consulting a doctor.

[0072] By analyzing the changing trends of blood pressure and related health data, the health risks that users may face, such as hypertension and cardiovascular disease, can be predicted in advance. Through early warning, users can take preventive measures before problems occur to avoid the occurrence or deterioration of diseases; by analyzing a large amount of user data, the model can assess the long-term health risks of individuals and provide more accurate prevention strategies. For example, some users may have a higher risk of cardiovascular disease, and the model can recommend that they take more active preventive measures.

[0073] The health reports and risk assessment results it generates can help doctors make more accurate diagnosis and treatment decisions. Especially when primary medical resources are limited, the analysis results provided by the model can serve as an important reference for doctors to optimize diagnosis and treatment plans. Through real-time monitoring and early warning, users can avoid frequent visits to the doctor due to minor health fluctuations, thereby saving medical resources and personal time. Only go to the hospital for professional examinations or treatment when necessary.

[0074] Through continuous monitoring and feedback, users' compliance with health management can be enhanced. For example, when a user adheres to a healthy lifestyle (such as regular exercise and taking medicine on time) for a period of time, the model can encourage them to maintain these behaviors through positive feedback or incentive mechanisms (such as health points). Personalized reminders can be set based on the user's daily behavior habits, such as taking medicine on time and exercising appropriately, to help users better manage their health.

[0075] By accumulating long-term data on users, we can discover health patterns of individuals or groups, and provide data support for subsequent health management and disease prevention. These data analysis results can also be used for medical research, public health strategy formulation, etc. We can predict the health status of users in the future, such as simulating the impact of different lifestyles on blood pressure, so as to provide more targeted health management suggestions.

[0076] It can also be connected to smart devices (such as smart bracelets, blood pressure monitors, etc.) to achieve remote real-time monitoring of user health data, which is particularly suitable for groups such as elderly people at home and patients with chronic diseases. Doctors can remotely understand the health status of patients through the real-time data provided by the model, reducing the need for face-to-face visits. Users can consult with doctors online through the health reports and data generated by the model to obtain professional advice and guidance. This model is particularly suitable for patients who cannot go to the hospital frequently, such as those living in remote areas.

[0077] The health data reports (such as weekly, monthly, quarterly, and annual lines) generated by the present invention can be intuitively displayed through charts and other means to help users understand their health status and changing trends more clearly. This visual feedback helps to improve users' health awareness and encourage them to participate more actively in health management. It not only provides data, but also provides health education in combination with analysis results, such as explaining the reasons for blood pressure fluctuations, the impact of exercise on health, etc., to help users better understand and manage their own health. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0079] Figure 1 It is a flow chart of the blood pressure artificial intelligence analysis system of the present invention. DETAILED DESCRIPTION

[0080] See also Figure 1 As shown, in this specific implementation, the 4G blood pressure monitor adopts the following technical solution: it includes a digital blood pressure detection module 1, a data communication module 2, and a display module 3. The blood pressure detection information of the digital blood pressure detection module 1 is connected to the 4G network or the wifi network through the data communication module 1 to connect to the smart device 4, and the blood pressure detection information displays relevant information through the display module 3; the smart device 4 records and analyzes blood pressure information, customer information, user mood information, exercise information, and medication information through a large data model and artificial intelligence, and generates through artificial intelligence: health reports, diet adjustment suggestions, exercise suggestions, and work and rest regularity suggestions.

[0081] In this specific implementation, the 4G sphygmomanometer artificial intelligence blood pressure analysis system adopts the following steps: 1. user data collection and entry, 2. user data storage and management, 3. user data analysis and modeling, 4. user blood pressure data report generation and health advice report, 5. user interaction and feedback, 6. model optimization and update;

[0082] The step 1 of collecting and entering user data includes: collecting the systolic pressure, diastolic pressure, pulse pressure difference and other data of the user's blood pressure in real time; recording the user's heart beats per minute and synchronizing it with the blood pressure data; manually inputting the user's mood information, and the user's mood can be set to several levels including: very happy, happy, normal, unhappy, and very unhappy; the user's exercise type, duration, intensity, and interval time before measuring blood pressure, the user's exercise information can be manually collected or automatically entered by smart devices such as bracelets; recording the user's medication use on the day, including the drug name, dosage, time, etc.

[0083] The user data storage and management in step 2 includes: designing special tables or data sets for different categories of data, such as blood pressure data tables, pulse data tables, mood data tables, etc.; cleaning and preprocessing the data, cleaning the collected data, removing abnormal values ​​and missing values, and standardizing the data format to ensure the consistency and accuracy of the data;

[0084] The data cleaning is as follows: 2.1. The processing of missing values ​​is as follows: 2.1.1. Delete missing values. If the missing value ratio is low, such as less than 5% of the data, delete these records directly to ensure the integrity of the data set; 2.1.2. Use interpolation to process missing values: For time series data, such as blood pressure and pulse, linear interpolation, spline interpolation and other methods can be used to fill in missing values.

[0085] The linear interpolation formula is:

[0086] Among them, x1 and x2 are the values ​​at time points t1 and t2 respectively, and t is the time point of the missing value;

[0087] 2.1.3. For non-time series data, use the mean or median of the same feature column to fill missing values ​​to reduce data bias;

[0088] 2.1.4 Use the K-nearest neighbor algorithm to fill missing values ​​based on the values ​​of similar data points, which is suitable for high-dimensional data;

[0089] 2.2. The method for handling outliers is as follows: 2.2.1 Statistical analysis: outliers are detected by box plot and Z score statistical method;

[0090] Z-score formula:

[0091] Where x is a data point, μ is the mean, and σ is the standard deviation. Generally, data points with a Z score greater than 3 or less than -3 can be considered outliers. 2.2.2. Delete outliers. For extreme outliers, delete them directly to avoid misleading the analysis results. 2.2.3. Replace outliers: Outliers can be replaced with adjacent values ​​or medians to smooth the data distribution. 2.2.4. Boundary processing: If the data has physical or logical boundaries (such as blood pressure data will not be lower than 30 mmHg or higher than 300 mmHg), the values ​​outside the boundaries will be adjusted to the boundary values.

[0092] 2.3. The method for handling duplicate data is as follows: 2.3.1. Identify duplicate records and check the duplicate records in the data set through primary keys or composite keys (such as timestamps, user IDs, etc.); 2.3.2. Delete duplicate records. If the duplicate records are exactly the same, delete only one and keep only one; if the data is slightly different, keep the newer or more complete data record.

[0093] The data preprocessing is specifically as follows:

[0094] 2.4. Data standardization, including 2.4.1. Min-Max standardization, which scales the data to a specific interval (such as [0,1]), is applicable to situations where the dimensions of the features are inconsistent;

[0095] The Min-Max normalization formula is:

[0096] 2.4.2. Z-score standardization, converting the data into a standard normal distribution with a mean of 0 and a standard deviation of 1;

[0097] The Z-score normalization formula is:

[0098] 2.5. Discretize the data, including: 2.5.1. Equal-width binning, dividing continuous data into intervals of equal width, and the number of data points in each interval may be different; 2.5.2. Equal-frequency binning, dividing the data into intervals of equal frequency, and the number of data points in each interval is the same; 2.5.3. Custom binning, manually define the binning intervals according to business needs or data distribution, such as dividing blood pressure levels into three intervals: low pressure, normal pressure, and high pressure.

[0099] 2.6. Smooth the data, including:

[0100] 2.6.1 Moving average, replaces the original data with the average value of data within a certain window to smooth short-term fluctuations, and is suitable for time series data;

[0101] The specific algorithm formula is:

[0102] 2.6.2. Exponentially Weighted Moving Average (EWMA), which gives higher weights to newer data points, is suitable for scenarios that respond quickly to trend changes;

[0103] The specific algorithm formula is: EWMA t =α×X t +(1-α)×EWMA t-1

[0104] Among them, α is the smoothing factor, X t For the current data point, EWMA t-1 is the smoothed value at the previous moment.

[0105] 2.7. Feature engineering, including feature selection: selecting features that have a greater impact on the target variable based on their importance (e.g., through correlation analysis, information gain, etc.); feature construction: generating new features by combining and transforming existing features, such as constructing a health score from blood pressure and pulse data; feature encoding: encoding categorical features, such as converting mood data into numerical form, using One-Hot Encoding or Label Encoding;

[0106] 2.8. Data verification, including: data distribution check, by drawing histograms, scatter plots, etc. to check data distribution and verify the effect of data cleaning and preprocessing; model performance evaluation, using part of the cleaned data for model training and testing, and evaluating the impact of data preprocessing on model performance;

[0107] 2.9. Storing and updating data, including data after cleaning: storing the cleaned and preprocessed data in a database or data warehouse for subsequent analysis and modeling; regular updating and maintenance: regularly cleaning and preprocessing new data to ensure the continued stability of data quality. Data cleaning and preprocessing in step 2 are the basic steps for data analysis and modeling. They ensure the quality of the data and lay a solid foundation for subsequent analysis and model training. Through these steps, data noise and errors can be minimized and the accuracy and reliability of the model can be improved.

[0108] The specific steps of step 3 user data analysis and modeling are as follows:

[0109] 3.1. Trend analysis: weekly, monthly, quarterly and annual blood pressure analysis. Specifically, the moving average method is used to smooth short-term fluctuations and show long-term trends.

[0110] The moving average method formula is:

[0111] Among them, P tis the blood pressure value on day t, and n is the time window size;

[0112] 3.2. Data aggregation: aggregate daily blood pressure, pulse, mood, exercise, and medication data, and calculate daily health scores through weighted average;

[0113] The weighted average health score formula is:

[0114] HealthScore=w1×BloodPressureScore+w2×PulseScore+w3×

[0115] MoodScore+w4×ExerciseScore+w5×MedicationScore

[0116] Among them, w1, w2, ..., w5 are the weights of each indicator, which can be adjusted through data training.

[0117] 3.3. Anomaly detection, using statistical analysis and machine learning algorithms (such as outlier detection algorithms) to identify abnormal changes in blood pressure, such as sudden increases or decreases;

[0118] Outlier detection formula:

[0119] Detect outliers by standard deviation and mean: Threshold = μ ± 2σ

[0120] Where μ is the mean value of blood pressure and σ is the standard deviation;

[0121] 3.4. User health risk assessment: Based on long-term data, combined with the user's historical data and health records, the user's health risk is assessed through algorithms such as logistic regression and random forest, and future blood pressure trends are predicted;

[0122] The logistic regression model formula is as follows:

[0123]

[0124] Where Y is the health risk result, X1, X2, X3…, X n is the input feature;

[0125] In the logistic regression model in step 3.4, X1, X2, X3…, X n Represents the model's feature variables or independent variables; these feature variables are extracted from user data and used to predict target variables, such as health risks; the sources of feature variables can include the following categories:

[0126] 3.4.1. Blood pressure data related features:

[0127] Systolic Pressure: It refers to the pressure of blood against the arterial wall when the heart contracts.

[0128] Diastolic Pressure: It indicates the pressure of blood against the arterial wall when the heart relaxes.

[0129] Pulse Pressure: The difference between systolic and diastolic blood pressure, used to assess arteriosclerosis and cardiovascular risk.

[0130] Mean Arterial Pressure (MAP): The weighted average of systolic and diastolic blood pressure, used to comprehensively evaluate blood pressure levels.

[0131] MAP formula:

[0132] 3.4.2 Pulse data related features:

[0133] Heart Rate: the number of heart beats per minute;

[0134] Heart Rate Variability (HRV): The variation in the interval between heart beats is an important indicator of heart health.

[0135] 3.4.3. Features related to user mood information;

[0136] Mood Score: the user's self-assessed mood status, such as a score of 1-10;

[0137] Mood Category: Mood label selected by the user, such as "Relaxed", "Tense", etc.

[0138] 3.4.4. Features related to motion information;

[0139] Exercise Duration: the total time the user spends exercising every day;

[0140] Exercise Intensity: The intensity level of exercise, such as mild, moderate, and severe;

[0141] Exercise Type: The type of exercise performed by the user, such as running, swimming, yoga, etc.

[0142] 3.4.5. Characteristics related to medication information;

[0143] Medication Type: the type of medication the user is taking;

[0144] Medication Dosage: The dosage of each medication;

[0145] Medication Timing: The time when you take medication may affect the effect of the drug on blood pressure.

[0146] 3.4.6. Time-related features;

[0147] Time Features: such as the time when the data was collected, which can include periodic features such as day, week, month, etc.

[0148] Seasonal Features: Take into account the impact of seasonal changes on blood pressure, such as the changing trends of blood pressure in summer and winter.

[0149] 3.4.7. Interaction Features

[0150] Feature combination: such as the interaction term between blood pressure and pulse, to assess the impact of the relationship between the two on health risks;

[0151] Polynomial Features: such as the square term of blood pressure and the square term of pulse, which capture nonlinear relationships.

[0152] 3.4.8. Historical data characteristics;

[0153] Historical blood pressure trend: blood pressure change trend over the past period of time, such as the average systolic blood pressure over the past 7 days;

[0154] Historical medication status: medication patterns over the past period of time, such as the number of times medication was taken in the past week.

[0155] 3.4.9. Individual characteristics;

[0156] Age: The user's age may significantly affect blood pressure and cardiovascular health;

[0157] Gender: Blood pressure and cardiovascular risk may differ between different genders;

[0158] Body Mass Index (BMI): The user's BMI value. Weight is closely related to health risks.

[0159] In the logistic regression model, X1, X2, X3, ..., X n These are features extracted from user data, which can cover multi-dimensional information such as blood pressure, pulse, mood, exercise, medication, etc. Through these features, the model can more accurately predict the user's health risks and generate personalized health recommendations.

[0160] The step 4 generates a user blood pressure data report and a health advice report, generates a blood pressure trend chart based on the weekly, monthly, quarterly and annual lines, and displays the user's health status graphically; generates a detailed health report based on the data analysis results, including the user's blood pressure trend analysis, heart health assessment, exercise and eating habits assessment; shares the data report and health report, which can be shared through social software or by adding social functions in the app itself;

[0161] The health advice report includes: Dietary advice: Based on the user's blood pressure and pulse data, it provides low-salt, low-fat, high-fiber dietary advice; Work and rest advice: Based on the user's mood and exercise data, it recommends a reasonable work and rest time and relaxation activities; Exercise advice: Based on the user's exercise data, it provides appropriate exercise intensity and type advice; Medication advice: Based on the user's medication records and blood pressure changes, it recommends medication adjustments or reminders.

[0162] The step 5 user interaction and feedback includes 5.1. Real-time monitoring: through mobile or web applications, users can view their blood pressure data and health reports in real time and get personalized suggestions; 5.2. User feedback mechanism: users can provide feedback on their physical condition or health information, and the model will adjust itself based on these feedbacks to improve the accuracy of the analysis.

[0163] The step 6 is the optimization and updating of the model; including 6.1. Continuous optimization of the machine learning model: as the amount of data increases, the model is continuously trained and optimized using new data to improve prediction accuracy; 6.2. Algorithm adjustment: based on user feedback and actual results, the parameters and weights in the algorithm are adjusted to adapt to individual differences and new health trends.

[0164] The continuous optimization and algorithm adjustment of the machine learning model in step 6.1 is an important step to improve the prediction accuracy and stability. The commonly used optimization methods and specific model information are as follows:

[0165] 6.1.1. Model selection and initial construction:

[0166] Model selection: In the initial stage, you may choose different models to try, such as Logistic Regression, Decision Tree, Random Forest, Support Vector Machine (SVM), Neural Network, etc.

[0167] Initial model training: Use the cleaned and preprocessed data set to perform initial training on the selected model and evaluate its basic performance (such as accuracy, precision, recall, etc.).

[0168] 6.1.2. Model performance evaluation and tuning are:

[0169] Cross-Validation: Use K-fold cross-validation (such as 5-fold or 10-fold) to evaluate the generalization ability of the model and reduce the risk of overfitting. Process: Divide the dataset into K subsets, use one of the subsets as the validation set in turn, and the remaining K-1 subsets as the training set, repeat K times, and finally take the average performance.

[0170] Hyperparameter Tuning: Use Grid Search or Random Search to tune the model’s hyperparameters and find the optimal parameter combination.

[0171] Common hyperparameters:

[0172] Logistic regression: Regularization strength (C);

[0173] Random Forest: number of decision trees (n_estimators), maximum tree depth (max_depth);

[0174] SVM: kernel function type (kernel), penalty coefficient (C);

[0175] Neural Networks: learning rate, number of hidden layers and nodes, activation function, etc.

[0176] 6.1.3. Feature engineering optimization is:

[0177] Feature Selection: Select features that have a greater impact on the target variable based on feature importance (such as through Lasso regression, recursive feature elimination (RFE), etc.), and eliminate irrelevant or redundant features.

[0178] Feature construction and transformation: Use polynomial features, interactive features, logarithmic transformation and other methods to enhance the expressiveness of the model and capture nonlinear relationships.

[0179] Dimensionality Reduction: Use PCA (Principal Component Analysis) or LDA (Linear Discriminant Analysis) and other methods to reduce the dimensionality of high-dimensional data to reduce the complexity of the model and the risk of overfitting.

[0180] 6.1.4. Model integration and improvement:

[0181] Ensemble Learning:

[0182] Combine the prediction results of multiple models to improve the robustness and prediction performance of the model.

[0183] The specific method is:

[0184] Bagging: Like random forest, the prediction results of multiple decision trees are averaged or voted.

[0185] Boosting: Such as XGBoost and LightGBM, gradually increase the weight of difficult-to-classify samples and build a strong classifier.

[0186] Stacking: Use the prediction results of multiple different models as new features and train the second layer model for the final prediction.

[0187] Model Blending: Use linear weighted averaging, logistic regression and other methods to fuse the output results of multiple models to enhance prediction capabilities.

[0188] The model fusion formula is:

[0189]

[0190] in, are the prediction results of different models, w1, w2, ..., w n is the corresponding weight.

[0191] 6.1.5. Model monitoring and feedback are adjusted to:

[0192] Model Monitoring: Continuously monitor the model’s predictive performance, especially on new data, to see if the model is experiencing performance degradation or increased bias;

[0193] Feedback Adjustment: Dynamically adjust the model’s weights, hyperparameters, or data processing methods based on the results of model monitoring to ensure that the model continues to maintain high performance;

[0194] Online learning: For real-time streaming data, use online learning algorithms (such as SGD and OnlineSVM) to gradually update the model to keep the model adaptable to the latest data;

[0195] Update the model parameters using gradient descent:

[0196] θ t+1 =θ t -η▽ θ L(θ t ;x t ,y t ).

[0197] 6.1.6. Algorithm adjustment and model update are as follows:

[0198] Retraining the model: When the data set expands significantly or the features change, it may be necessary to retrain or update the existing model; the specific retraining methods are: Incremental Learning: gradually update the weights based on the existing model; Periodic retraining: regularly retrain the model with the latest data set to adapt to new situations.

[0199] Algorithm upgrade: With the development of technology and the increase in data volume, more complex models or algorithms (such as deep learning and reinforcement learning) may be needed to improve predictive capabilities.

[0200] 6.1.7. Model validation and evaluation: Model validation: Use the validation set or test set to evaluate the final performance of the model. Common indicators include AUC, F1-score, precision, recall, etc. Model deployment and launch: Deploy the final optimized model to the production environment to provide real-time or batch prediction services.

[0201] The continuous optimization and algorithm adjustment of machine learning models is an iterative process, including model selection, feature engineering, ensemble learning, hyperparameter tuning, feedback adjustment, etc. Through these methods, the prediction performance of the model can be continuously improved to ensure the accuracy and reliability of the model in practical applications.

[0202] The working principle of the present invention is as follows: it uses a home-use digital blood pressure measuring instrument to record the measurement results and simultaneously record the pulse, user's mood, exercise status, medication status, etc., to form daily, weekly, monthly, quarterly, and annual blood pressure reports, etc., and uses artificial intelligence analysis of a large model to give corresponding health reports, health advice, and medication and medical advice, etc., and can use corresponding software to publish and share blood pressure information of relatives and friends, and use the software's own social functions to care for the blood pressure status of relatives and friends.

[0203] After adopting the above technical solution, the beneficial effects of the present invention are: it provides practical advantages in personalized health management, prevention of chronic diseases, optimization of medical resources, support for telemedicine, etc. These advantages not only improve the health management level of users, but also promote the efficient use of medical resources and promote the intelligent and personalized development of health management.

[0204] The above description is only used to illustrate the technical solution of the present invention rather than to limit it. Other modifications or equivalent substitutions made to the technical solution of the present invention by ordinary technicians in this field should be included in the scope of the claims of the present invention as long as they do not depart from the spirit and scope of the technical solution of the present invention.

Claims

1. 4G sphygmomanometer, characterized by: It includes a digital blood pressure detection module 1, a data communication module 2, and a display module 3. The blood pressure detection information of the digital blood pressure detection module 1 is connected to a smart device 4 through a 4G network or a wifi network via the data communication module 1, and the blood pressure detection information displays relevant information through the display module 3; the smart device 4 records and analyzes blood pressure information, customer information, user mood information, exercise information, and medication information through a large data model and artificial intelligence, and generates through artificial intelligence: health reports, diet adjustment suggestions, exercise suggestions, and work and rest regularity suggestions.

2. Artificial intelligence blood pressure analysis system, characterized by: It adopts the following steps:

1. User data collection and entry, 2. User data storage and management, 3. User data analysis and modeling, 4. User blood pressure data report generation and health advice report, 5. User interaction and feedback, 6. Model optimization and update; The step 1 of collecting and entering user data includes: collecting the systolic pressure, diastolic pressure, pulse pressure difference and other data of the user's blood pressure in real time; recording the user's heart beats per minute and synchronizing it with the blood pressure data; manually inputting the user's mood information, and the user's mood can be set to several levels including: very happy, happy, normal, unhappy, and very unhappy; the user's exercise type, duration, intensity, and interval time before measuring blood pressure, the user's exercise information can be manually collected or automatically entered by smart devices such as bracelets; recording the user's medication use on the day, including the drug name, dosage, time, etc.

3. The artificial intelligence blood pressure analysis system according to claim 2, characterized in that: The step 2 of user data storage and management includes: designing special tables or data sets for different categories of data, such as blood pressure data tables, pulse data tables, mood data tables, etc.; cleaning and preprocessing the data, cleaning the collected data, removing outliers and missing values, and standardizing the data format to ensure the consistency and accuracy of the data.

4. The artificial intelligence blood pressure analysis system according to claim 3, characterized in that: The data cleaning is specifically as follows: 2.

1. The specific processing of missing values ​​is as follows: 2.1.

1. Delete missing values. If the missing value ratio is low, such as less than 5% of the data, delete these records directly to ensure the integrity of the data set; 2.1.

2. Use interpolation to handle missing values: For time series data, such as blood pressure and pulse, linear interpolation, spline interpolation and other methods can be used to fill in missing values The linear interpolation formula is: in, x1 and x2 are the values ​​at time points t1 and t2, respectively, and t is the time point of the missing value; 2.1.

3. For non-time series data, use the mean or median of the same feature column to fill missing values ​​to reduce data bias; 2.1.4 Use the K-nearest neighbor algorithm to fill missing values ​​based on the values ​​of similar data points, which is suitable for high-dimensional data; 2.

2. The method for dealing with outliers is: 2.2.1 Statistical analysis, outliers were detected by box plot and Z score statistical method; Z-score formula: in, x is the data point, μ is the mean, and σ is the standard deviation; usually, data points with a Z score greater than 3 or less than -3 can be considered outliers; 2.2.

2. Delete outliers. For extreme outliers, delete them directly to avoid misleading the analysis results. 2.2.

3. Replace outliers: Outliers can be replaced with adjacent values ​​or medians to smooth data distribution; 2.2.

4. Boundary processing: If the data has physical or logical boundaries (such as blood pressure data will not be lower than 30mmHg or higher than 300mmHg), the values ​​outside the boundaries will be adjusted to the boundary values; 2.

3. The method for processing duplicate data is as follows: 2.3.

1. Identify duplicate records and check the duplicate records in the data set by primary key or composite key (such as timestamp, user ID, etc.); 2.3.

2. Delete duplicate records. If the duplicate records are exactly the same, delete only one and keep only one; if the data is slightly different, keep the newer or more complete data record; The data preprocessing is specifically as follows: 2.

4. Data standardization, including 2.4.

1. Min-Max standardization, which scales the data to a specific interval (such as [0,1]), is applicable to situations where the dimensions of the features are inconsistent; The Min-Max normalization formula is: 2.4.

2. Z-score standardization, converting the data into a standard normal distribution with a mean of 0 and a standard deviation of 1; The Z-score normalization formula is: 2.

5. Discretize the data, including: 2.5.

1. Equal-width binning: divide continuous data into equal-width intervals, and the number of data points in each interval may be different; 2.5.

2. Equal frequency binning, divide the data into equal frequency intervals, and the number of data points in each interval is the same; 2.5.

3. Customized binning: manually define binning intervals based on business needs or data distribution, such as dividing blood pressure levels into three intervals: low pressure, normal pressure, and high pressure; 2.

6. Smooth the data, including: 2.6.1 Moving average, replaces the original data with the average value of data within a certain window to smooth short-term fluctuations, and is suitable for time series data; The specific algorithm formula is: 2.6.

2. Exponentially Weighted Moving Average (EWMA), which gives higher weights to newer data points, is suitable for scenarios that respond quickly to trend changes; The specific algorithm formula is: EWMA t =α×X t +(1-α)×EWMA t-1 in, α is the smoothing factor, X t For the current data point, EWMA t-1 is the smoothed value of the previous moment; 2.

7. Feature Engineering, including Feature Selection: Select features that have a greater impact on the target variable based on their importance (e.g., through correlation analysis, information gain, etc.); Feature construction: Generate new features by combining and transforming existing features, such as constructing a health score through blood pressure and pulse data; Feature encoding: Encode categorical features, such as converting mood data into numerical form, using One-Hot Encoding or Label Encoding; 2.

8. Data verification, including: Data distribution check: Check data distribution by drawing histograms, scatter plots, etc. to verify the effect of data cleaning and preprocessing; Model performance evaluation: Use some cleaned data for model training and testing to evaluate the impact of data preprocessing on model performance; 2.

9. Storing and updating data, including data after cleaning: storing the cleaned and preprocessed data in a database or data warehouse for subsequent analysis and modeling; regular updating and maintenance: regularly cleaning and preprocessing new data to ensure the continued stability of data quality; data cleaning and preprocessing in step 2 are basic steps for data analysis and modeling, which ensures the quality of the data and lays a solid foundation for subsequent analysis and model training. Through these steps, data noise and errors can be minimized and the accuracy and reliability of the model can be improved.

5. The artificial intelligence blood pressure analysis system according to claim 2, characterized in that: The specific steps of step 3 user data analysis and modeling are as follows: 3.

1. Trend analysis: weekly, monthly, quarterly and annual blood pressure analysis. Specifically, the moving average method is used to smooth short-term fluctuations and show long-term trends. The moving average method formula is: in, P t is the blood pressure value on day t, and n is the time window size; 3.

2. Data aggregation: aggregate daily blood pressure, pulse, mood, exercise, and medication data, and calculate daily health scores through weighted average; The weighted average health score formula is: HealthScore=w1×BloodPressureScore+w2×PulseScore+w3× MoodScore+w4×ExerciseScore+w5×MedicationScore Among them, w1, w2, ..., w5 are the weights of each indicator, which can be adjusted through data training. 3.

3. Anomaly detection, using statistical analysis and machine learning algorithms (such as outlier detection algorithms) to identify abnormal changes in blood pressure, such as sudden increases or decreases; Outlier detection formula: Detect outliers by standard deviation and mean: Threshold = μ ± 2σ Where μ is the mean value of blood pressure and σ is the standard deviation; 3.

4. User health risk assessment: Based on long-term data, combined with the user's historical data and health records, the user's health risk is assessed through algorithms such as logistic regression and random forest, and future blood pressure trends are predicted; The logistic regression model formula is as follows: Where Y is the health risk result, X1, X2, X3…, X n is the input feature.

6. The artificial intelligence blood pressure analysis system according to claim 5, characterized in that: In the logistic regression model in step 3.4, X1, X2, X3…, X n Represents the characteristic variable or independent variable of the model; These feature variables are extracted from user data and used to predict target variables, such as health risks; the sources of feature variables can include the following categories: 3.4.

1. Blood pressure data related features: Systolic Pressure: It indicates the pressure of blood against the arterial wall when the heart contracts. Diastolic Pressure: It indicates the pressure of blood on the arterial wall when the heart relaxes. Pulse Pressure: The difference between systolic and diastolic blood pressure, used to assess arteriosclerosis and cardiovascular risk; Mean Arterial Pressure (MAP): The weighted average of systolic and diastolic blood pressure, used to comprehensively evaluate blood pressure levels; MAP formula: 3.4.2 Pulse data related features: Heart Rate: the number of heart beats per minute; Heart Rate Variability (HRV): The variation in heartbeat intervals, an important indicator of heart health; 3.4.

3. Features related to user mood information; Mood Score: the user's self-assessed mood status, such as a score of 1-10; Mood Category: The mood label selected by the user, such as "relaxed", "tense", etc. 3.4.

4. Features related to motion information; Exercise Duration: the total time the user spends exercising every day; Exercise Intensity: The intensity level of exercise, such as mild, moderate, and severe; Exercise Type: the type of exercise performed by the user, such as running, swimming, yoga, etc. 3.4.

5. Characteristics related to medication information; Medication Type: the type of medication the user is taking; Medication Dosage: The dosage of each medication; Medication Timing: The time of taking medication may affect the effect of the drug on blood pressure; 3.4.

6. Time-related features; Time Features: such as the time when the data was collected, which can include periodic features such as day, week, month, etc. Seasonal Features: Taking into account the impact of seasonal changes on blood pressure, such as the changing trends of blood pressure in summer and winter; 3.4.

7. Interaction Features Feature combination: such as the interaction term between blood pressure and pulse, to assess the impact of the relationship between the two on health risks; Polynomial Features: such as the square term of blood pressure and the square term of pulse, which capture nonlinear relationships; 3.4.

8. Historical data characteristics; Historical blood pressure Trend: The trend of blood pressure changes over the past period of time, such as the average systolic blood pressure over the past 7 days; Medication history: medication patterns over the past period of time, such as the number of times medication was taken in the past week; 3.4.

9. Individual characteristics; Age: The user's age may significantly affect blood pressure and cardiovascular health; Gender: Blood pressure and cardiovascular risk may differ between different genders; Body Mass Index (BMI): The user's BMI value. Weight is closely related to health risks; In the logistic regression model, X1, X2, X3, ..., X n These are features extracted from user data. These features can cover multi-dimensional information such as blood pressure, pulse, mood, exercise, and medication. Through these features, the model can more accurately predict the user's health risks and generate personalized health recommendations. The step 4 generates a user blood pressure data report and a health advice report, generates a blood pressure trend chart based on the weekly, monthly, quarterly and annual lines, and displays the user's health status graphically; generates a detailed health report based on the data analysis results, including the user's blood pressure trend analysis, heart health assessment, exercise and eating habits assessment; shares the data report and health report, which can be shared through social software or by adding social functions in the app itself; The health advice report includes: Dietary advice: Based on the user's blood pressure and pulse data, it provides low-salt, low-fat, high-fiber dietary advice; Work and rest advice: Based on the user's mood and exercise data, it recommends a reasonable work and rest time and relaxation activities; Exercise advice: Based on the user's exercise data, it provides appropriate exercise intensity and type advice; Medication advice: Based on the user's medication records and blood pressure changes, it recommends medication adjustments or reminders.

7. The artificial intelligence blood pressure analysis system according to claim 2, characterized in that: The step 5 user interaction and feedback includes 5.

1. Real-time monitoring: through the mobile or web application, users can view their blood pressure data and health reports in real time and obtain personalized suggestions; 5.

2. User feedback mechanism: Users can provide feedback on their physical condition or health information, and the model will adjust itself based on this feedback to improve the accuracy of the analysis.

8. The artificial intelligence blood pressure analysis system according to claim 2, characterized in that: The step 6, model optimization and updating, includes 6.

1. continuous optimization of the machine learning model: as the amount of data increases, the model is continuously trained and optimized using new data to improve prediction accuracy; 6.

2. Algorithm adjustment: According to user feedback and actual results, adjust the parameters and weights in the algorithm to adapt to individual differences and new health trends; The continuous optimization and algorithm adjustment of the machine learning model in step 6.1 is an important step to improve the prediction accuracy and stability. The commonly used optimization methods and specific model information are as follows: 6.1.

1. Model selection and initial construction: Model selection: In the initial stage, you may choose different models to try, such as Logistic Regression, Decision Tree, Random Forest, Support Vector Machine (SVM), Neural Network, etc. Initial model training: Use the cleaned and preprocessed data set to perform initial training on the selected model and evaluate its basic performance (such as accuracy, precision, recall, etc.); 6.1.

2. Model performance evaluation and tuning are: Cross-Validation: Use K-fold cross-validation (such as 5-fold or 10-fold) to evaluate the generalization ability of the model and reduce the risk of overfitting; process : Divide the dataset into K subsets, use one of the subsets as the validation set in turn, and the remaining K-1 subsets as the training set, repeat K times, and finally take the average performance; Hyperparameter Tuning: Use Grid Search or Random Search to tune the model's hyperparameters and find the optimal parameter combination. Common hyperparameters: Logistic regression: Regularization strength (C); Random Forest: number of decision trees (n_estimators), maximum tree depth (max_depth); SVM: kernel function type (kernel), penalty coefficient (C); Neural networks: learning rate, number of hidden layers and nodes, activation function, etc. 6.1.

3. Feature engineering optimization is: Feature Selection: Select features that have a greater impact on the target variable based on feature importance (such as through Lasso regression, recursive feature elimination (RFE), etc.), and eliminate irrelevant or redundant features; Feature construction and transformation: Use polynomial features, interactive features, logarithmic transformation and other methods to enhance the model's expressiveness and capture nonlinear relationships; Dimensionality Reduction: Use PCA (principal component analysis) or LDA (linear discriminant analysis) to reduce the dimensionality of high-dimensional data to reduce the complexity of the model and the risk of overfitting; 6.1.

4. Model integration and improvement: Ensemble Learning: Combine the prediction results of multiple models to improve the robustness and prediction performance of the model; The specific method is: Bagging: Like random forest, the prediction results of multiple decision trees are averaged or voted; Boosting: such as XGBoost and LightGBM, which gradually increase the weight of difficult-to-classify samples and build a strong classifier; Stacking: Use the prediction results of multiple different models as new features and train the second layer model for the final prediction; Model Blending: Use linear weighted averaging, logistic regression and other methods to fuse the output results of multiple models to enhance prediction capabilities; The model fusion formula is: in, are the prediction results of different models, w1, w2, ..., w n is the corresponding weight; 6.1.

5. Model monitoring and feedback are adjusted to: Model Monitoring: Continuously monitor the model’s predictive performance, especially on new data, to see if the model is experiencing performance degradation or increased bias; Feedback Adjustment: Dynamically adjust the model’s weights, hyperparameters, or data processing methods based on the results of model monitoring to ensure that the model continues to maintain high performance; Online learning: For real-time streaming data, use online learning algorithms (such as SGD and Online SVM) to gradually update the model to keep the model adaptable to the latest data; Update the model parameters using gradient descent: 6.1.

6. Algorithm adjustment and model update are as follows: Retraining models: When the dataset expands significantly or the features change, it may be necessary to retrain or update the existing model. The specific retraining methods are: Incremental Learning: gradually update the weights based on the existing model; Periodic retraining: regularly retrain the model with the latest dataset to adapt to new situations; Algorithm upgrade: With the development of technology and the increase in data volume, more complex models or algorithms (such as deep learning and reinforcement learning) may be needed to improve prediction capabilities; 6.1.

7. Model validation and evaluation: Model validation: Use the validation set or test set to evaluate the final performance of the model. Common indicators include AUC, F1-score, precision, recall, etc. Model deployment and launch: Deploy the final optimized model to the production environment to provide real-time or batch prediction services.

Citation Information

Cited By

  • Blood pressure interpretation processing method and device based on ambulatory blood pressure monitoring and related equipment

    CN119724574A

  • Interrogation guide system based on interactive data acquisition

    CN120145171A

  • An inquiry guidance system based on interactive data collection

    CN120145171B

  • Engineering risk prediction method based on artificial intelligence

    CN120355216A

  • Hypertensive patient behavior guidance generation device, equipment, storage medium and system

    CN120581129A