A weight loss trajectory prediction method, system, device and storage medium

Abstract

This invention discloses a method, system, device, and storage medium for predicting weight loss trajectories, belonging to the field of health management technology. The method includes: acquiring basic physiological data of users and collecting multi-dimensional feature data, including at least two categories of gut health features, rhythm and compliance features, hydration status features, refined behavioral features, metabolic monitoring features, and dynamic body composition features; constructing an initial weight loss trajectory prediction model based on the multi-dimensional feature data; dynamically updating model parameters using a Bayesian fusion strategy; and identifying weight loss plateau states and outputting early warning information through a Hidden Markov Model. This method, by fusing multi-dimensional core features and using a Bayesian fusion strategy for dynamic parameter updates, achieves improved prediction accuracy, early warning of plateau periods, and quantitative support for intervention plans. The predicted RMSE for weight loss after 3 months is reduced to 0.65 kg, and the recall rate for plateau warnings reaches 92%.

Description

Technical Field

[0002] This invention relates to the field of health management technology, and in particular to a method, system, device, and storage medium for predicting weight loss trajectory. Background Technology

[0004] With the global obese population continuing to grow, scientific and personalized weight loss management has become a core issue in the field of public health. Current mainstream weight loss management technologies can be mainly divided into three categories, all of which have unavoidable core flaws.

[0005] Static, experience-based predictive solutions, such as apps like Mint Health and NOOM, rely solely on fixed calorie calculation formulas and general rules of thumb, completely ignoring individual metabolic differences. Their feature dimensions are extremely limited, only including basic indicators such as weight and total calories, without covering the core physiological and behavioral characteristics that affect weight loss. They also lack the ability to warn of plateaus and have significant long-term predictive biases.

[0006] Patent application CN115035981A, entitled "A Weight Prediction Method Based on Deep Learning," discloses a weight prediction method using a deep learning black-box model. This method lacks physiological interpretability and has poor user compliance; it does not incorporate core influencing factors such as urinary ketone levels, gut health, sleep rhythm, and dynamic changes in body composition; it can only identify factors retrospectively during plateau periods and cannot provide early warnings; and the intervention plan lacks quantitative support.

[0007] Simplified group rule-based management solutions, such as the standardized management systems of various offline weight loss institutions, are based on average group data and fixed rule guidance, resulting in extremely low personalization. They completely ignore individual characteristics such as behavioral compliance, intestinal metabolism, and hydration status. When a user's weight loss trajectory deviates significantly from the group's pattern, the predictive effect becomes completely ineffective.

[0008] In summary, existing technologies generally suffer from the following core problems: severe lack of feature dimensions, complete neglect of core metabolic influencing factors; inadequate plateau management capabilities; weak dynamic learning capabilities of the model; lack of physiological interpretability, resulting in low user compliance; and intervention plans lack quantitative support, leading to poor feasibility. Summary of the Invention

[0010] This invention addresses the technical problems of existing technologies, such as single feature dimensions, insufficient plateau management capabilities, weak dynamic learning capabilities of models, and poor interpretability, by collecting multi-dimensional feature data, dynamically updating model parameters using a Bayesian fusion strategy, and achieving early warning of plateau periods through a hidden Markov model. It also produces technical effects such as improved prediction accuracy, early warning of plateau periods, and quantitative support for intervention schemes.

[0011] To achieve the above objectives, the present invention adopts the following technical solution:

[0012] A weight loss trajectory prediction method includes the following steps: acquiring basic physiological data of users; constructing an initial prediction model based on the basic physiological data; wherein, collecting multi-dimensional feature data of users, the multi-dimensional feature data includes at least two categories of gut health features, rhythm and compliance features, hydration status features, refined behavioral features, metabolic monitoring features, and dynamic body composition features; dynamically updating the parameters of the initial prediction model based on the multi-dimensional feature data; identifying weight loss plateau states based on a hidden Markov model and outputting early warning information; the dynamic update adopts a Bayesian fusion strategy, calculating the posterior distribution of model parameters based on the user's historical data, and the posterior distribution is used to update the parameters of the initial prediction model; the hidden states of the hidden Markov model include rapid weight loss period, slow weight loss period, plateau period, and weight recovery period, and calculating the probability of entering the plateau period in the future through a forward algorithm, outputting an early warning when the probability exceeds a preset threshold.

[0013] Through the above technical solutions, the present invention achieves the technical effects of significantly improving prediction accuracy, providing early warning of plateau periods, and providing quantitative support for intervention plans.

[0014] Furthermore, the gut health characteristics include at least one of daily bowel movement frequency and stool consistency. This setup enables the capture of core physiological signals affecting weight loss efficiency at the gut metabolic level, enhancing the sensitivity of the predictive model to individual metabolic differences.

[0015] Furthermore, the rhythm and compliance characteristics include at least one of daily sleep onset time, sleep duration, and weight loss check-in frequency. This setting enables the quantification of the impact of user behavior patterns on weight loss effectiveness, enhancing the model's responsiveness to behavioral interventions.

[0016] Furthermore, the hydration status characteristics include at least one of daily water intake and water intake compliance rate. This setup enables the monitoring of key hydration factors affecting metabolic efficiency and weight, improving the accuracy of predicting short-term weight fluctuations.

[0017] Furthermore, the refined behavioral characteristics include at least one of the following: details of daily food intake (three meals a day) and daily exercise. This setup enables the precise capture of dynamic changes in daily energy intake and expenditure, supporting the generation of personalized intervention programs.

[0018] Furthermore, the metabolic monitoring features include daily morning urine ketone body measurements. This setup enables real-time monitoring of fat metabolism, providing direct physiological evidence for weight loss effectiveness assessment and intervention adjustments.

[0019] Furthermore, the dynamic body composition characteristics include at least one of skeletal muscle mass, body fat percentage, body fat percentile, and lean body mass. This setup allows for the differentiation between fat loss and muscle loss during weight loss, improving the scientific rigor of weight loss quality assessment.

[0020] Furthermore, the basic physiological data includes at least one of sex, age, height, initial weight, initial BMI, target weight, history of underlying medical conditions, and resting basal metabolic rate. This setup enables the construction of personalized initial predictive models, providing accurate baseline parameters for subsequent dynamic updates.

[0021] Furthermore, the initial prediction model is a differential equation model based on energy balance theory, expressed as: W(t) = W∞ + (W0 - W∞) × exp(-k×t), where W(t) is the predicted weight at time t, W0 is the initial weight, W∞ is the steady-state weight, and k is the weight loss rate parameter. This setting gives the prediction model clear physiological significance, enhancing its interpretability and user trust.

[0022] Furthermore, the initial values ​​of the steady-state body weight and weight loss rate parameters are predicted using a pre-trained gradient boosting tree regression model. This setup enables the rapid acquisition of individualized initial parameter estimates using population data, shortening the model warm-up time.

[0023] Furthermore, the Bayesian fusion strategy includes: calculating the prior distribution of model parameters based on multi-dimensional feature data of the group; calculating the likelihood function of parameters based on recent dynamic feature data of users; calculating the posterior distribution of parameters according to the Bayesian formula; and fusion of prior and posterior parameters weighted by data volume and attendance compliance. This setup allows the model parameters to be continuously optimized as user data accumulates, taking into account both group experience and individual differences.

[0024] Furthermore, the update formula for the weighted fusion is: μ_new = w × μ_ML + (1-w) × μ_prior, where μ_ML is the maximum likelihood estimate based on user dynamic data, μ_prior is the prior parameter estimate, and w is the weight coefficient. This setting provides a clear mathematical expression for the parameter update process, ensuring the traceability and adjustability of the fusion strategy.

[0025] Furthermore, the weighting coefficient ranges from 0 to 1, dynamically increasing with the amount and frequency of user check-in data. The formula for calculating the weighting coefficient is: w = min(1, α × N / N_threshold × β × F / F_threshold), where N is the amount of user check-in data, N_threshold is the baseline data volume threshold, F is the check-in frequency, F_threshold is the baseline frequency threshold, and α and β are adjustment coefficients, with α + β = 1. This setting allows the model to adaptively adjust the contribution ratio between group experience and individual data, gradually increasing the degree of personalization as user data accumulates.

[0026] Furthermore, the criteria for determining a plateau are: a weight change of less than 0.1 kg over 7 consecutive days with an average daily decrease of less than 0.02 kg, excluding short-term weight stagnation caused by water retention. This setting provides an objective and quantifiable standard for determining a plateau, avoiding errors and misjudgments based on subjective assessment.

[0027] Furthermore, this includes using an accelerated failure time model to predict the duration of the plateau phase, with at least three of the following covariates: diet control rate, exercise achievement rate, sleep time compliance, water intake achievement rate, bowel regularity, and check-in continuity. This setup allows for the quantitative prediction of the plateau phase's duration, enabling the setting of reasonable psychological expectations and intervention timing for users.

[0028] Furthermore, this includes constructing multiple sets of virtual intervention feature vectors, each corresponding to a different intervention scheme. These vectors are then input into an accelerated failure time model to obtain the predicted plateau duration under the corresponding intervention scheme, and the optimal intervention scheme with the shortest plateau breakthrough time is selected. This setup enables the simulation and comparison of the effects of various intervention strategies, providing users with data-driven personalized intervention recommendations.

[0029] Furthermore, this includes: constructing user feature vectors based on multi-dimensional user feature data; calculating the similarity between the current user and users in the historical user database using a weighted distance algorithm; retrieving similar user cases; and enhancing the interpretability of the prediction results based on similar user cases. This setup allows the prediction results to be corroborated by similar real-world cases, increasing users' trust in the prediction conclusions and intervention plans.

[0030] Furthermore, gut health features were assigned a weight of 1.5, dynamic body composition features a weight of 1.3, and other features a weight of 1.0. This setting allows the model to highlight the core features that have the greatest impact on weight loss, improving prediction accuracy and the relevance of similar case matching.

[0031] Furthermore, it also includes visualizing at least one of the predicted weight loss trajectory, plateau warning information, and intervention plans. This setup allows complex prediction results to be presented in an intuitive form, lowering the barrier to user understanding and improving user compliance.

[0032] Secondly, this invention provides a weight loss trajectory prediction system, including a data acquisition module, a data preprocessing module, a model building module, a parameter update module, a plateau warning module, an intervention simulation module, and an interactive display module. The data acquisition module collects multi-dimensional feature data from users; the data preprocessing module preprocesses the collected feature data; the model building module constructs an initial weight loss trajectory prediction model; the parameter update module dynamically updates the model parameters using a Bayesian fusion strategy; the plateau warning module identifies weight loss plateaus based on a Hidden Markov Model and outputs a warning; the intervention simulation module predicts the duration of the plateau and simulates the effects of intervention programs; and the interactive display module visually presents the prediction results to users.

[0033] Through the above technical solution, the present invention realizes a systematic and modular weight loss trajectory prediction capability, with each functional module working together to support a complete prediction, early warning and intervention closed loop.

[0034] Furthermore, the data acquisition module includes a basic physiological data acquisition unit and a multi-dimensional feature acquisition unit. The basic physiological data acquisition unit is used to collect the user's basic physiological data; the multi-dimensional feature acquisition unit is used to collect the user's gut health characteristics, rhythm and compliance characteristics, hydration status characteristics, refined behavioral characteristics, metabolic monitoring characteristics, and dynamic body composition characteristics. This setup gives the data acquisition function a clear modular structure, facilitating system expansion and maintenance.

[0035] Furthermore, the parameter update module includes a prior distribution calculation unit, a likelihood function calculation unit, a posterior distribution calculation unit, and a parameter fusion unit. The prior distribution calculation unit calculates the prior distribution of the model parameters; the likelihood function calculation unit calculates the likelihood function of the parameters; the posterior distribution calculation unit calculates the posterior distribution of the parameters according to Bayes' theorem; and the parameter fusion unit weightedly fuses the prior and posterior parameters to update the model. This setup provides a clear functional division for the implementation of the Bayesian fusion strategy, ensuring the accuracy and traceability of parameter updates.

[0036] Furthermore, the plateau phase early warning module includes a hidden state identification unit, a probability calculation unit, and an early warning output unit. The hidden state identification unit is used to identify the hidden states of the hidden Markov model; the probability calculation unit is used to calculate the probability of entering the plateau phase using a forward algorithm; and the early warning output unit is used to output early warning information when the probability exceeds a preset threshold. This setup enables the plateau phase early warning function to have a complete processing chain, from state identification to probability calculation to early warning output, ensuring the timeliness and accuracy of the early warning.

[0037] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-mentioned weight loss trajectory prediction method.

[0038] Through the above technical solution, the present invention realizes the hardware carrier of the weight reduction trajectory prediction method, enabling the method to be deployed and run on various computing devices.

[0039] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described weight reduction trajectory prediction method.

[0040] Through the above technical solution, the present invention realizes the storage and distribution of the weight reduction trajectory prediction method, which facilitates the dissemination and application of the method.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] 1. Significantly improved prediction accuracy: For the first time, core features such as gut health, sleep rhythm, compliance, hydration status, refined behavior, and dynamic body composition are incorporated into the weight loss prediction model. The RMSE of weight prediction after 3 months is reduced to 0.65kg, which is significantly better than existing technologies.

[0043] 2. Early warning of plateau period: By using a hidden Markov model that integrates features from all dimensions, the recall rate of early warning of plateau period is 92% and the precision rate is 89% by 1 to 2 weeks, which can accurately identify the risk of plateau period.

[0044] 3. Quantitative support for intervention programs: By accelerating the failure time model to predict the duration of the plateau period, the accuracy of MAE is reduced to 2.1 days, which can quantify the effect of multi-dimensional intervention programs.

[0045] 4. Improved long-term prediction stability: Based on the Bayesian fusion strategy, the model weights are dynamically adjusted by combining user check-in frequency and data volume, which solves the problem of decreased long-term prediction accuracy caused by static parameters.

[0046] 5. Enhanced interpretability: Based on weighted similarity case matching with full-dimensional features, it accurately matches real cases that closely match the user's physiology, behavior, and weight loss rhythm, enhancing the user's trust in the prediction results and intervention plan. Attached Figure Description

[0048] Figure 1 This is a schematic flowchart of the weight reduction trajectory prediction method according to Embodiment 1 of the present invention;

[0049] Figure 2 This is a schematic diagram of multi-dimensional feature data acquisition in Embodiment 1 of the present invention;

[0050] Figure 3 This is a schematic diagram of the Bayesian fusion strategy parameter update process in Embodiment 1 of the present invention;

[0051] Figure 4 This is a schematic diagram of the hidden Markov model plateau early warning architecture in Embodiment 1 of the present invention;

[0052] Figure 5 This is a schematic diagram of the simulation process of the accelerated failure time model intervention effect in Embodiment 1 of the present invention;

[0053] Figure 6 This is a schematic diagram of the similar user case matching process in Embodiment 1 of the present invention;

[0054] Figure 7 This is a schematic diagram of the system architecture and data flow of Embodiment 1 of the present invention.

[0055] In the picture:

[0056] 10-Data Acquisition Module; 11-Basic Physiological Data Acquisition Unit; 12-Multi-Dimensional Feature Acquisition Unit; 101-Daily Bowel Movement Frequency; 102-Stool Characteristics; 103-Probiotic Use; 201-Sleep Onset Time; 202-Sleep Duration; 203-Weight Loss Check-in Frequency; 204-Check-in Continuity; 301-Daily Water Intake; 302-Water Intake Compliance Rate; 401-Dietary Intake Details; 402-Daily Exercise; 501-Urine Ketone Body Measurement; 601-Skeletal Muscle Mass; 602-Body Fat Percentage; 603-Lean Body Mass; 701-Rest Basal Metabolic Rate; 20-Data Preprocessing Module; 21-Outlier Removal Unit; 22-Normalization Unit; 23-Missing Value Imputation Unit; 30-Model Building Module; 31-Differential Equation Model Building Unit; 32-Gradient Boosting Tree Regression Model Unit; W∞-Stability Weight Parameter; k- Weight loss rate parameter; 40 - Parameter update module; 41 - Prior distribution calculation unit; 42 - Likelihood function calculation unit; 43 - Posterior distribution calculation unit; 44 - Parameter fusion unit; μ_prior - Prior parameter estimate; μ_ML - Maximum likelihood estimate; μ_new - Updated parameter value; w - Weight coefficient; 50 - Plateau warning module; 51 - Hidden state identification unit; 52 - Probability calculation unit; 53 - Warning output unit; S1 - Rapid weight loss period; S2 - Slow weight loss period; S3 - Plateau period; S4 - Weight recovery period; 60 - Intervention simulation module; 61 - Accelerated failure time model unit; 62 - Virtual intervention vector construction unit; 63 - Program selection unit; 70 - Interactive display module; 71 - Trajectory curve display unit; 72 - Warning information display unit; 73 - Intervention program display unit; 74 - Similar case display unit. Detailed Implementation

[0058] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments. It should be noted that, without conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments.

[0059] In the description of this invention, it should be noted that directional terms such as "center," "lateral," "longitudinal," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise" indicate the orientation and positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. They should not be construed as limiting the specific protection scope of this invention.

[0060] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features. Thus, the use of "first" and "second" to define a feature may explicitly or implicitly include one or more of that feature, and in the description of this invention, "a number" means two or more, unless otherwise explicitly specified.

[0061] In this invention, unless otherwise explicitly specified and limited, terms such as "set" and "install" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can also refer to a mechanical connection; they can refer to a direct connection or a connection through an intermediate medium; or they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0062] Example 1

[0063] Example Overview: This example uses a 32-year-old female user who is trying to lose weight to illustrate the complete implementation process of the weight loss trajectory prediction method.

[0064] Reference Figure 1 and Figure 2 The weight loss trajectory prediction method in this embodiment includes the following steps:

[0065] Step S1: Collect full-dimensional feature data.

[0066] The data acquisition module 10 collects multi-dimensional feature data of users.

[0067] The user's basic physiological data were collected: female, 32 years old, 162cm tall, initial weight 68kg, initial BMI 25.9, target weight 55kg, no underlying diseases, and resting basal metabolic rate of 1320kcal / day.

[0068] Intestinal health characteristics were collected: daily bowel movement frequency 101 was 1 time, stool characteristics 102 were normal, and probiotic use 103 was supplementation with Bifidobacterium triple live bacteria capsules, 2 capsules twice a day.

[0069] Collect rhythm and compliance characteristics: daily sleep time 201 is 00:30-01:00, sleep duration 202 is 6-7 hours, weight loss check-in frequency 203 is committed to daily check-in, and check-in continuity 204 records actual check-in status.

[0070] Hydration status characteristics were collected: the daily water intake for 301 was 1000ml / day, and the water intake compliance rate for 302 was 55.6%, which did not reach the recommended intake of 1800ml / day.

[0071] Collected detailed behavioral characteristics: Dietary intake details 401 show that the three meals are mainly refined carbohydrates, with protein accounting for about 12%, and the average daily calorie intake is 1800kcal; Daily exercise 402 shows that there is no regular exercise in daily life, and the average daily extra exercise consumption is 0kcal.

[0072] Metabolic monitoring features collected: The daily morning urine ketone body measurement value of 501 is the user's daily self-test data.

[0073] Dynamic body composition characteristics were collected: initial skeletal muscle mass 601 was 21.3 kg, initial body fat percentage 602 was 32%, and lean body mass 603 was 45.2 kg.

[0074] Step S2: Construction of the initial weight reduction trajectory prediction model.

[0075] The model building module 30 constructs an initial weight loss trajectory prediction model based on the collected multi-dimensional feature data.

[0076] Reference Figure 3 Differential equation model building unit 31 constructs a differential equation model based on energy balance theory, with the expression: W(t) = W∞ + (W0 - W∞) × exp(-k×t).

[0077] Gradient boosting tree regression model unit 32 is based on a pre-trained gradient boosting tree regression model. It takes seven categories of full-dimensional feature data as input and predicts the initial parameters: steady-state weight parameter W∞ is 54.2 kg, and weight loss rate parameter k is 0.012.

[0078] The initial weight loss trajectory is generated based on the modified differential equation model: according to the formula W(t) = 54.2 + (68 -54.2) × exp(-0.012×t), the weight is expected to drop to 64.1kg in the 4th week (28 days), to 61.3kg in the 8th week (56 days), and to 59.2kg in the 12th week (84 days).

[0079] Step S3: Dynamically update model parameters.

[0080] The parameter update module 40 collects users' weight, diet, exercise, sleep time, water intake, bowel movements, and attendance data daily, and updates body composition data monthly.

[0081] Reference Figure 3 The prior distribution calculation unit 41 calculates the prior distribution p(θ) of the model parameter θ = [W∞, k] based on the full-dimensional feature data of a large-scale population.

[0082] The likelihood function calculation unit 42 calculates the likelihood function L(Data|θ) of the parameters based on the user's dynamic feature data over the past n days.

[0083] The posterior distribution calculation unit 43 calculates the posterior distribution of the parameters p(θ|Data) ∝ L(Data|θ) × p(θ) according to Bayes' formula.

[0084] The parameter fusion unit 44 fused prior and posterior parameters by weighting data volume and attendance compliance, and the update formula was: μnew = w × μML + (1-w) × μ_prior.

[0085] The formula for calculating the weighting coefficient w is: w = min(1, α × N / Nthreshold × β × F / Fthreshold), where N is the amount of user check-in data, Nthreshold is the baseline data amount threshold set to 14 days, F is the check-in frequency, Fthreshold is the baseline frequency threshold set to 100%, and α and β are adjustment coefficients. In this embodiment, α = 0.5 and β = 0.5.

[0086] After a user checks in for 14 consecutive days, the weighting coefficient w = min(1, 0.5 × 14 / 14 × 0.5 × 1.0 / 1.0) = min(1, 0.5 × 1 × 0.5 × 1) = 0.25. Combined with the user's actual weight data, the updated weight loss rate parameter k is 0.011, the steady-state weight parameter W∞ is updated to 54.5kg, and the weight loss prediction trajectory is corrected accordingly.

[0087] The weighting coefficient w increases dynamically as the amount of user check-in data increases and the frequency of check-in increases: in the initial stage w=0.1, w increases to 0.5 after 30 consecutive days of check-in, and w increases to 0.8 after 90 consecutive days of check-in.

[0088] Step S4: Weight loss trajectory prediction and early warning of plateau period.

[0089] The plateau warning module 50, based on the updated personalized dynamic weight loss prediction model, predicts the user's weight for the next 1 to 12 weeks and generates a visual weight loss trajectory curve.

[0090] Reference Figure 4 The hidden state identification unit 51 defines the hidden state space of the hidden Markov model: S1 is the rapid weight loss period, S2 is the slow weight loss period, S3 is the plateau period, and S4 is the weight recovery period.

[0091] The probability calculation unit 52 calculates the probability of entering the plateau state S3 within the next 7 to 14 days based on the user's observation sequence over the past 14 days using a forward algorithm.

[0092] When the probability exceeds the preset threshold of 0.7, the early warning output unit 53 issues an early warning of the plateau period to the user and at the same time alerts the high-risk factors that may cause the plateau period.

[0093] The criteria for determining a plateau are: a weight change of less than 0.1 kg over 7 consecutive days and an average daily decrease of less than 0.02 kg, while excluding short-term weight stagnation caused by water fluctuations, to avoid misjudging weight stagnation caused by muscle growth as a plateau.

[0094] After a user checks in for 14 consecutive days, the HMM model calculates that the probability of entering a plateau period in the next 10 days is 78%, which exceeds the warning threshold of 0.7. The system immediately issues a plateau warning to the user, indicating high-risk factors such as insufficient protein intake, inadequate water intake, and going to bed too late.

[0095] Step S5: Prediction of plateau duration and simulation of intervention.

[0096] The intervention simulation module 60 uses an accelerated failure time model to predict the duration of the plateau period and quantifies the effects of different intervention schemes.

[0097] Reference Figure 5 Accelerated failure time model unit 61 constructs an AFT model, expressed as: ln(T) = β0 + β1X1 + β2X2 + ... + β p X p + σε, where T is the duration of the plateau period (number of days), and X is a covariate vector covering at least three of the following: diet control rate, exercise achievement rate, sleep time compliance, water intake achievement rate, bowel regularity, probiotic adherence, check-in continuity, muscle mass change rate, and body fat percentage change rate.

[0098] Virtual intervention vector construction unit 62 constructs multiple sets of virtual intervention feature vectors, each corresponding to a different intervention scheme.

[0099] The scheme selection unit 63 inputs each virtual intervention vector into the accelerated failure time model to obtain the predicted duration of the plateau period under the corresponding intervention scheme, and selects the optimal intervention scheme with the shortest plateau period breakthrough time.

[0100] Based on the AFT model, if users maintain their current behavior, the plateau period is expected to last 32 days.

[0101] Four virtual intervention schemes were constructed, and the simulation results are as follows:

[0102] Option A: Maintain current behavior, expected to last 32 days;

[0103] Option B: Increase protein intake to 25% + meet daily water intake requirements, expected to last for 18 days;

[0104] Option C: Option B + 30 minutes of aerobic exercise daily + bedtime earlier at 11:00 PM, expected to last 12 days;

[0105] Option D: Option C + Skip dinner every other day, expected to last for 10 days.

[0106] Select the optimal solution D and recommend it to the user.

[0107] Step S6: Matching similar user cases.

[0108] The interactive display module 70 retrieves similar user cases based on full-dimensional feature vectors, enhancing the interpretability of the prediction results.

[0109] Reference Figure 6 User feature vectors are constructed based on users' multi-dimensional feature data. The gut health feature is given a weight of 1.5 times, the dynamic body composition feature is given a weight of 1.3 times, and other features are given a weight of 1.0 times.

[0110] A weighted distance algorithm is used to calculate the similarity between the current user and users in the historical user database, and the five user cases with the highest similarity are retrieved from the historical user database.

[0111] Extract the actual weight loss trajectory of the case, the time of the plateau, the duration, the intervention plan used, and the breakthrough effect, and present it to the current user.

[0112] In this embodiment, five highly similar user cases were retrieved. Four of them entered a plateau in the third week of weight loss. Through the intervention measures of Plan D, they broke through the plateau in an average of 11 days and lost an average of 12.6 kg in 3 months.

[0113] Step S7: Visualize the prediction results.

[0114] The interactive display module 70 uses an interactive interface to visually display to users the predicted weight loss trajectory curve, plateau warning information, expected duration of the plateau, optimal intervention plan, and similar user cases.

[0115] Reference Figure 7 The trajectory curve display unit 71 displays the corrected weight loss trajectory curve; the early warning information display unit 72 displays early warning notices for the plateau period; the intervention plan display unit 73 displays the optimal intervention plan; and the similar case display unit 74 displays similar user cases and their effects.

[0116] Example 2

[0117] To verify the effectiveness of this invention, a control experiment was conducted.

[0118] Experimental group: 100 weight loss users, using the method of this invention for full-cycle weight loss prediction and management.

[0119] Control group 1: 100 weight loss users, using the traditional static prediction method (Mint Health APP).

[0120] Control group 2: 100 weight loss users, using a pure machine learning weight prediction method (patent CN115035981A).

[0121] Experiment duration: 3 months.

[0122] The evaluation indicators and experimental results are shown in the table below:

[0123] Evaluation indicators experimental group Control group 1 Control group 2 3-month weight prediction RMSE (kg) 0.65 1.45 0.95 Plateau period warning recall rate 92% No early warning capability 72% Plateau period early warning accuracy 89% No early warning capability 78% Plateau period warning F1 value 0.90 No early warning capability 0.75 Plateau period duration prediction (MAE in days) 2.1 No predictive ability 5.8 Average weight loss (kg) over 3 months 9.8 4.2 6.5

[0124] Experimental results show that the present invention is significantly superior to existing technologies in terms of weight prediction accuracy, plateau warning capability, and intervention effect, verifying the effectiveness and advancement of the present invention.

[0125] The above description is only a specific embodiment of the present invention, but the technical features of the present invention are not limited thereto. Any changes or modifications made by those skilled in the art within the scope of the present invention are covered by the patent scope of the present invention.

Claims

1. A method for predicting a weight loss trajectory, comprising the following steps: Obtain basic physiological data from users; The initial prediction model is constructed based on the aforementioned basic physiological data, characterized by the following steps: collecting multi-dimensional feature data of users, including at least two categories of gut health features, rhythm and compliance features, hydration status features, refined behavioral features, metabolic monitoring features, and dynamic body composition features; dynamically updating the parameters of the initial prediction model based on the multi-dimensional feature data; identifying weight loss plateau states based on a hidden Markov model and outputting early warning information; wherein, the dynamic update adopts a Bayesian fusion strategy, calculating the posterior distribution of model parameters based on the user's historical data, and the posterior distribution is used to update the parameters of the initial prediction model; the latent states of the hidden Markov model include rapid weight loss period, slow weight loss period, plateau period, and weight recovery period, and calculating the probability of entering the plateau period in the future through a forward algorithm, outputting an early warning when the probability exceeds a preset threshold.

2. The weight loss trajectory prediction method according to claim 1, characterized in that, The multidimensional feature data includes: gut health features, including daily bowel movement frequency, stool characteristics, and / or probiotic consumption; rhythm and compliance features, including daily sleep onset time, sleep duration, weight loss check-in frequency, and / or check-in continuity; hydration status features, including daily water intake and / or water intake target achievement rate; refined behavioral features, including details of daily food intake and / or daily exercise; metabolic monitoring features, including daily morning urine ketone body measurements; and dynamic body composition features, including skeletal muscle mass, body fat percentage, body fat percentage, and / or lean body mass.

3. The weight loss trajectory prediction method according to claim 1, characterized in that, The initial prediction model is a differential equation model based on energy balance theory. The expression of the differential equation model is: W(t) = W∞ + (W0 - W∞)× exp(-k×t), where W(t) is the predicted weight at time t, W0 is the initial weight, W∞ is the steady-state weight, and k is the weight loss rate parameter. The initial values ​​of the steady-state weight W∞ and the weight loss rate parameter k are predicted by a pre-trained gradient boosting tree regression model.

4. The weight loss trajectory prediction method according to claim 1, characterized in that, The Bayesian fusion strategy includes: calculating the prior distribution of model parameters based on multi-dimensional feature data of the group; calculating the likelihood function of parameters based on recent dynamic feature data of users; calculating the posterior distribution of parameters according to the Bayesian formula; and fusion of prior and posterior parameters by weighting data volume and attendance compliance. The update formula for weighted fusion is: μ_new = w × μ_ML + (1-w) × μ_prior, where μ_ML is the maximum likelihood estimate based on user dynamic data, μ_prior is the prior parameter estimate, and w is the weight coefficient.

5. The weight loss trajectory prediction method according to claim 4, characterized in that, The weighting coefficient w ranges from 0 to 1 and increases dynamically with the increase of user check-in data volume and check-in frequency. The formula for calculating the weighting coefficient w is: w = min(1, α × N / N_threshold × β × F / F_threshold), where N is the user check-in data volume, N_threshold is the baseline data volume threshold, F is the check-in frequency, F_threshold is the baseline frequency threshold, α and β are adjustment coefficients, and α + β = 1.

6. The weight loss trajectory prediction method according to claim 1, characterized in that, Also includes: The accelerated failure time model is used to predict the duration of the plateau phase, with at least three of the following as covariates: diet control rate, exercise target achievement rate, sleep time compliance, water intake target achievement rate, bowel regularity, and attendance continuity. Multiple sets of virtual intervention feature vectors are constructed, each corresponding to a different intervention scheme. The accelerated failure time model is then input to obtain the predicted duration of the plateau phase under the corresponding intervention scheme, and the optimal intervention scheme with the shortest plateau phase breakthrough time is selected.

7. A weight loss trajectory prediction system, characterized in that, include: The data acquisition module (10) is used to collect multi-dimensional feature data of users; The data preprocessing module (20) is used to preprocess the collected feature data; The model building module (30) is used to build an initial weight loss trajectory prediction model; the parameter update module (40) is used to dynamically update the model parameters using a Bayesian fusion strategy; the plateau warning module (50) is used to identify the weight loss plateau based on the hidden Markov model and output a warning; the intervention simulation module (60) is used to predict the duration of the plateau and simulate the effect of the intervention plan; and the interactive display module (70) is used to visualize the prediction results to the user.

8. The weight loss trajectory prediction system according to claim 7, characterized in that, The parameter update module (40) includes: a prior distribution calculation unit (41) for calculating the prior distribution of model parameters; a likelihood function calculation unit (42) for calculating the likelihood function of the parameters; a posterior distribution calculation unit (43) for calculating the posterior distribution of the parameters according to Bayes' theorem; and a parameter fusion unit (44) for weighted fusion of prior and posterior parameters to update the model.

9. The weight reduction trajectory prediction system according to claim 7, characterized in that, The plateau phase early warning module (50) includes: a hidden state identification unit (51) for identifying the hidden state of the hidden Markov model; a probability calculation unit (52) for calculating the probability of entering the plateau phase through a forward algorithm; and an early warning output unit (53) for outputting early warning information when the probability exceeds a preset threshold.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the weight loss trajectory prediction method as described in any one of claims 1 to 6; or, a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the weight loss trajectory prediction method as described in any one of claims 1 to 6.

Images