An ai-enabled doping risk automated identification system and method
By using AI technology for data collection, feature construction, and the XGBoost classification model, the problems of resource waste and misjudgment in existing doping tests have been solved, enabling automated and accurate identification and management of athletes' doping risks.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-16
- Publication Date
- 2026-07-10
AI Technical Summary
Existing doping control strategies suffer from resource waste and false negatives, and are unable to effectively identify high-risk athletes. Traditional statistical methods are ineffective in identifying high-risk athletes in extremely imbalanced data.
By employing AI technology, the system acquires multi-dimensional data through a data acquisition module, calculates feature vectors through a feature construction module, performs automatic reasoning using an XGBoost classification model, and outputs risk labels in conjunction with a risk decision-making module, thereby achieving accurate identification of doping risks for athletes.
It has improved the accuracy and targeting of doping risk identification, reduced the waste of testing resources, and achieved automated and precise doping risk management.
Smart Images
Figure CN122369938A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical and health information technology, and in particular to an AI-based automatic doping risk identification system and method. Background Technology
[0002] Currently, the World Anti-Doping Agency (WADA) and national anti-doping centers generally adopt a comprehensive doping control strategy. This strategy involves conducting regular or random tests on all registered athletes at a uniform frequency, without distinguishing between athletes' sport types, historical violations, behavioral patterns, or performance fluctuations. For example, the China Anti-Doping Agency conducts tens of thousands of doping tests annually, covering all registered athletes. This strategy has significant technical problems: a large amount of testing resources are allocated to low-risk athletes with a long history of no violations, compliant behavior, and stable performance, while high-risk athletes fail to receive differentiated and focused monitoring, leading to diminishing marginal utility of testing resources.
[0003] In existing doping risk assessment technologies, inspectors primarily rely on human experience to independently analyze single-dimensional signals such as athletes' whereabouts reports, performance fluctuations, and historical violations. For example, when an athlete experiences a sudden surge in performance, the inspector may list them as a high-priority focus; when an athlete's whereabouts reports are ambiguous or frequently changed, the inspector may increase the frequency of checks. However, these methods lack the ability to systematically quantify and integrate multi-source heterogeneous data, including athlete basic information (such as sport type and coach's violation history), whereabouts information (such as update frequency and address accuracy), doping control data (such as test coverage and regularity of test intervals), and outcome management data (such as past violations and suspension records). Human experience-based judgment suffers from strong subjectivity, inconsistent standards, and difficulty in handling high-dimensional data, easily leading to missed detections or misjudgments.
[0004] Furthermore, the proportion of doping violations among all athletes is extremely low, typically less than one in a thousand. Traditional statistical methods (such as logistic regression) tend to predict all samples as risk-free when dealing with such extremely imbalanced data, failing to effectively identify the few violators. Therefore, this invention proposes an AI-based automatic doping risk identification system and method to address the problems existing in the prior art. Summary of the Invention
[0005] To address the aforementioned problems, the present invention aims to propose an AI-based automatic doping risk identification system and method. This invention achieves quantitative fusion and automatic reasoning of multi-dimensional heterogeneous data, overcomes the subjectivity of human experience judgment and the limitations of single signal analysis, and significantly improves the accuracy and relevance of doping risk identification, thereby solving the problems existing in the prior art.
[0006] To achieve the objectives of this invention, the invention is implemented through the following technical solution: an AI-based automatic doping risk identification system and method, comprising the following modules: The data acquisition module is used to acquire athletes' basic information, whereabouts, doping test results, and outcome management data. The feature construction module, connected to the data acquisition module, is used to calculate the performance mutation index and coach violation mark based on athlete basic information data, calculate the ambiguity of whereabouts and the abnormality of whereabouts update frequency based on whereabouts information data, calculate the inspection coverage rate, the proportion of inspections during abnormal periods, and the irregularity of inspection intervals based on doping inspection data, and calculate the number of past violations and the number of days from the end of the suspension to the current date based on the results management data. Then, the performance mutation index, coach violation mark, whereabouts ambiguity, abnormality of whereabouts update frequency, inspection coverage rate, proportion of inspections during abnormal periods, irregularity of inspection intervals, number of past violations, and the number of days from the end of the suspension to the current date are combined into a feature vector and output. The AI inference module, connected to the feature construction module, is used to calculate the output feature vector through the trained XGBoost classification model and output the risk probability value. The risk decision module, connected to the AI inference module, is used to compare the risk probability value with a preset risk threshold. When the risk probability value is greater than or equal to the preset risk threshold, a risky label is output; when the risk probability value is less than the preset risk threshold, a risk-free label is output.
[0007] Further improvements include: the athlete's basic information data includes a sequence of past competition results, the type of sport participated in, dietary habits, and a coach's history of doping violations; the location information data includes the grid precision value of the location reporting address, the location reporting timestamp sequence, and the number of times the location was temporarily changed; the doping test data includes the test timestamp, test type, and a report flag, wherein the test types include blood tests and dried blood spot tests; and the result management data includes a history of doping violations and a record of penalties, wherein the record of penalties includes the start date and end date of the ban.
[0008] Further improvements are made in the following ways: the performance mutation index is the standard score of the most recent competition result relative to the average of an individual's historical results, and the coach violation indicator is a binary variable indicating whether the coach has a history of doping violations.
[0009] The further improvement lies in the following: the fuzziness of the location is the numerical value corresponding to the grid precision value of the location reporting address, the anomaly of the location update frequency is the coefficient of variation of the location reporting timestamp interval, and the coefficient of variation is the ratio of the standard deviation of the location reporting timestamp interval to the mean of the location reporting timestamp interval.
[0010] Further improvements are made in the following ways: the inspection coverage rate is the ratio of the actual number of inspections in the past year to the expected number of inspections set according to the project type; the abnormal period inspection ratio is the ratio of the number of inspections during nighttime or holiday periods to the total number of inspections; and the irregularity of the inspection interval is the standard deviation of the time interval between two adjacent inspections in the past year.
[0011] A further improvement is that the XGBoost classification model is trained using a cost-sensitive learning strategy, where the ratio of the weight of positive samples to the weight of negative samples is equal to the ratio of the number of negative samples to the number of positive samples in the training set. An AI-based method for automatically identifying doping risks includes the following steps: Step 1: Obtain athletes' basic information, whereabouts information, doping test results, and outcome management data; Step 2: Calculate the performance mutation index and coach violation mark based on the athlete's basic information data; calculate the ambiguity of whereabouts and the abnormality of whereabouts update frequency based on the whereabouts information data; calculate the inspection coverage rate, the proportion of inspections during abnormal periods, and the irregularity of inspection intervals based on the doping inspection data; calculate the number of past violations and the number of days from the end of the suspension to the current date based on the results management data; and then combine the performance mutation index, coach violation mark, whereabouts ambiguity, abnormality of whereabouts update frequency, inspection coverage rate, proportion of inspections during abnormal periods, irregularity of inspection intervals, number of past violations, and the number of days from the end of the suspension to the current date into a feature vector. Step 3: Input the feature vector into the pre-trained XGBoost classification model and output the risk probability value; Step 4: Compare the risk probability value with the preset risk threshold. When the risk probability value is greater than or equal to the preset risk threshold, output a risky label. When the risk probability value is less than the preset risk threshold, output a risk-free label.
[0012] A further improvement is that, in step two, the time span of the tracking update frequency anomaly index is the past 365 days, which is the time span of the tracking report timestamp interval sequence.
[0013] A further improvement is made in step three, where the training data of the XGBoost classification model is divided into a training set, a validation set, and a test set in chronological order. The training set is used for learning model parameters, the validation set is used for hyperparameter tuning, and the test set is used for final performance evaluation.
[0014] A further improvement is made in step four, where the preset risk threshold is 0.5.
[0015] The beneficial effects of this invention are as follows: (1) This invention utilizes the collaborative work of a data acquisition module, a feature construction module, an AI reasoning module, and a risk decision-making module to perform multi-dimensional fusion and quantitative analysis of athletes' basic information, whereabouts information, inspection status, and result management data. The XGBoost classification model outputs risk probability values and risk-free or risk-free labels. This technical solution overcomes the subjectivity of human experience judgment and the resource waste of full-coverage inspections, achieving automatic and accurate identification of doping risks. While maintaining the violation detection rate, it significantly reduces the workload of inspections and reduces the consumption of manpower, time, and testing reagents.
[0016] (2) The present invention sets different expected inspection times according to project type, adapts to the risk distribution characteristics of different projects, realizes full-process automated processing, does not depend on specific inspection equipment, and has strong portability. Attached Figure Description
[0017] Figure 1 This is a schematic diagram of the connection of the automatic risk identification system module of the present invention.
[0018] Figure 2 This is a schematic diagram of the automatic risk identification method of the present invention. Detailed Implementation
[0019] To enhance understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.
[0020] Example 1 according to Figure 1 As shown, this embodiment provides an AI-based automatic doping risk identification system, including the following modules: Data acquisition module: used to acquire athletes' basic information, whereabouts information, doping test results, and results management data.
[0021] Feature Construction Module: Connected to the data acquisition module, this module calculates the performance mutation index and coach violation flag based on athlete basic information data; calculates the ambiguity of whereabouts and the abnormality of whereabouts update frequency based on whereabouts information data; calculates the inspection coverage rate, the proportion of inspections during abnormal periods, and the irregularity of inspection intervals based on doping inspection data; and calculates the number of past violations and the number of days from the end of the suspension to the current date based on the results management data. Finally, it combines the performance mutation index, coach violation flag, whereabouts ambiguity, abnormality of whereabouts update frequency, inspection coverage rate, proportion of inspections during abnormal periods, irregularity of inspection intervals, number of past violations, and the number of days from the end of the suspension to the current date into a feature vector and outputs it.
[0022] AI Inference Module: Connected to the feature construction module, it is used to calculate the risk probability value by using the trained XGBoost classification model to output the feature vector.
[0023] Risk Decision Module: Connected to the AI Inference Module, it compares the risk probability value with a preset risk threshold. When the risk probability value is greater than or equal to the preset risk threshold, it outputs a risky label; when the risk probability value is less than the preset risk threshold, it outputs a risk-free label.
[0024] Example 2 according to Figure 2 As shown, this embodiment provides an automatic doping risk identification method using AI, including the following steps: Step 1: Obtain basic athlete information, whereabouts information, doping test results, and outcome management data.
[0025] Step 2: Calculate the performance mutation index and coach violation mark based on the athlete's basic information data; calculate the ambiguity of the whereabouts and the abnormality of the whereabouts update frequency based on the whereabouts information data; calculate the inspection coverage rate, the proportion of inspections during abnormal periods, and the irregularity of inspection intervals based on the doping inspection data; calculate the number of past violations and the number of days between the end of the suspension and the current date based on the results management data; and then combine the above items into a feature vector.
[0026] Step 3: Input the feature vector into the pre-trained XGBoost classification model and output the risk probability value.
[0027] Step 4: Compare the risk probability value with the preset risk threshold. When the risk probability value is greater than or equal to the preset risk threshold, output a risky label. When the risk probability value is less than the preset risk threshold, output a risk-free label.
[0028] Specifically, the athlete's basic information data includes a sequence of past competition results, the type of sport participated in, dietary habits, and a coach's history of doping violations. The whereabouts information data includes the grid precision value of the whereabouts declaration address, the sequence of whereabouts declaration timestamps, and the number of times the whereabouts were temporarily changed. The doping test data includes the test timestamp, the type of test, and the report status, among which the test types include blood tests and dried blood spot tests. The results management data includes a history of doping violations and records of penalties, among which records of penalties include the start date and end date of the ban.
[0029] Then: Performance Mutation Index: This is a standardized score of the most recent competition result relative to the average of an individual's historical results. The formula for its calculation is as follows: Performance mutation index = (most recent competition result - μ) / σ In the formula, μ is the arithmetic mean of all the athlete's competition results in the past two years, and σ is the sample standard deviation of all the athlete's competition results in the past two years.
[0030] Coach Violation Indicator: A binary variable indicating whether a coach has a history of doping violations. If the athlete's current coach or a coach who has coached the athlete within the past three years has been found to have committed a doping violation by an anti-doping agency, the coach violation indicator is set to 1; otherwise, it is set to 0.
[0031] Location Ambiguity: This is the numerical value corresponding to the grid precision of the location reporting address. The grid precision value is defined as follows: if the reported address contains the complete street name and house number or GPS coordinates, the precision value is 0; if the reported address only contains the city or region name, the precision value is 1; if the reported address only contains the provincial administrative region name, the precision value is 2.
[0032] Anomaly of location update frequency: This is the coefficient of variation of the interval between location reporting timestamps, and its calculation formula is as follows: Location update frequency anomaly score = (sd(Δt) / mean(Δt)) In the formula, Δt is the time interval (in days) between two consecutive timestamps of the athlete's whereabouts report in the past 365 days, sd(Δt) is the sample standard deviation of all intervals, and mean(Δt) is the arithmetic mean of all intervals.
[0033] Inspection coverage: This is the ratio of the actual number of inspections conducted in the past year to the expected number of inspections set by project type. The expected number of inspections set by project type is 6 times per year for physical fitness projects, 3 times per year for skill-based projects, and 4 times per year for mixed projects.
[0034] Abnormal inspection period ratio: This is the ratio of the number of inspections conducted during nighttime or holiday periods to the total number of inspections. Nighttime periods are defined as 22:00 to 06:00 the following day. Holiday periods are defined as national statutory holidays, and the calculation formula is as follows: The percentage of inspections during abnormal periods = (number of inspections at night + number of inspections on holidays) / total number of inspections.
[0035] Inspection interval irregularity: This is the standard deviation of the time interval between two consecutive inspections within the past year, calculated using the following formula: Check the irregularity of the interval = sd(Δτ) In the formula, Δτ is the time stamp interval (in days) between two consecutive doping tests of an athlete in the past year, and sd(Δτ) is the sample standard deviation of all intervals.
[0036] Previous violations: The total number of doping violations confirmed by the anti-doping agency for the athlete in the past five years.
[0037] Number of days from the end of the suspension to the current date: If the athlete has a suspension record, the number of days from the end of the most recent suspension to the current date is calculated; if the athlete has no suspension record, this value is 0.
[0038] For the XGBoost classification model, a cost-sensitive learning strategy is used for training. The training dataset consists of historical samples, each containing the aforementioned feature vector and its corresponding true label (risky = 1, no-risk = 0). The ratio of positive (risky) samples to negative (no-risk) samples in the training set is 1:1000. In the cost-sensitive learning strategy, the ratio of positive sample weights to negative sample weights is equal to the ratio of the number of negative samples to the number of positive samples in the training set, i.e., positive sample weight = 1000, negative sample weight = 1. The model is trained using GridSearchCV for hyperparameter tuning, and the final model parameters are: max_depth = 6, learning_rate = 0.1, n_estimators = 200.
[0039] The training data is divided into training set, validation set and test set in chronological order. The training set is used for model parameter learning, the validation set is used for hyperparameter tuning, and the test set is used for final performance evaluation.
[0040] Taking athlete A as an example, this athlete specializes in weightlifting (physical fitness). Their past two years' performance results are: 380kg, 385kg, 378kg, 382kg, 430kg. Their most recent competition result was 430kg. Their historical average performance μ = 381.25kg, and their standard deviation σ = 3.86kg. The performance mutation index = (430 - 381.25) / 3.86 = 12.63.
[0041] Coach violation mark: Athlete A's coach was found to have violated doping regulations in 2019, therefore the coach violation mark = 1.
[0042] Location Information: Athlete A reported his / her location 120 times in the past 365 days. All reported locations were at the city level with precision (grid precision value = 1), therefore the location ambiguity is 1. The mean of the location report timestamp interval sequence Δt is 3.04 days, the standard deviation is 2.17 days, and the location update frequency anomaly is 2.17 / 3.04 = 0.714.
[0043] Doping test results: Two tests were conducted in the past year (the expected number of tests for physical fitness events is 6), resulting in a test coverage rate of 2 / 6 = 0.333. One of the tests occurred at night (11:00 PM), representing an abnormal time period of 1 / 2 = 0.5. The time stamp interval between the two tests was 180 days, and the irregularity of the test interval was the standard deviation of 180 days (the standard deviation is calculated as 0 when there are only two data points; only the 180-day interval was actually recorded).
[0044] Results management data: Previous violations = 0, Number of days from the end of the suspension = 0.
[0045] Feature vector: {12.63, 1, 1, 0.714, 0.333, 0.5, 180, 0, 0}.
[0046] The feature vector was input into the trained XGBoost classification model, which output a risk probability value of p=0.87. With a preset risk threshold of 0.5, p≥0.5, the system outputs a "risky" label. Based on this, the anti-doping agency conducted a targeted test on athlete A, and the result confirmed a positive result.
[0047] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the present invention without departing from its framework and scope of application, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.
Claims
1. An automatic doping risk identification system using AI, characterized in that: Includes the following modules: The data acquisition module is used to acquire athletes' basic information, whereabouts, doping test results, and outcome management data. The feature construction module, connected to the data acquisition module, is used to calculate the performance mutation index and coach violation mark based on athlete basic information data, calculate the ambiguity of whereabouts and the abnormality of whereabouts update frequency based on whereabouts information data, calculate the inspection coverage rate, the proportion of inspections during abnormal periods, and the irregularity of inspection intervals based on doping inspection data, and calculate the number of past violations and the number of days from the end of the suspension to the current date based on the results management data. Then, the performance mutation index, coach violation mark, whereabouts ambiguity, abnormality of whereabouts update frequency, inspection coverage rate, proportion of inspections during abnormal periods, irregularity of inspection intervals, number of past violations, and the number of days from the end of the suspension to the current date are combined into a feature vector and output. The AI inference module, connected to the feature construction module, is used to calculate the output feature vector through the trained XGBoost classification model and output the risk probability value. The risk decision module, connected to the AI inference module, is used to compare the risk probability value with a preset risk threshold. When the risk probability value is greater than or equal to the preset risk threshold, a risky label is output; when the risk probability value is less than the preset risk threshold, a risk-free label is output.
2. The AI-based automatic doping risk identification system according to claim 1, characterized in that: The athlete's basic information data includes a sequence of past competition results, type of sport, dietary habits, and coach's history of doping violations. The location information data includes the grid precision value of the location reporting address, the location reporting timestamp sequence, and the number of times the location was temporarily changed. The doping test data includes the test timestamp, test type, and report flag, wherein the test types include blood tests and dried blood spot tests. The result management data includes history of doping violations and penalty records, wherein the penalty records include the start date and end date of the ban.
3. The AI-based automatic doping risk identification system according to claim 1, characterized in that: The performance mutation index is the standard score of the most recent competition result relative to the average of an individual's historical results, and the coach violation indicator is a binary variable indicating whether a coach has a history of doping violations.
4. The AI-based automatic doping risk identification system according to claim 1, characterized in that: The fuzziness of the location is the numerical value corresponding to the grid precision value of the location reporting address, and the anomaly of the location update frequency is the coefficient of variation of the location reporting timestamp interval. The coefficient of variation is the ratio of the standard deviation of the location reporting timestamp interval to the mean of the location reporting timestamp interval.
5. The AI-based automatic doping risk identification system according to claim 1, characterized in that: The inspection coverage rate is the ratio of the actual number of inspections in the past year to the expected number of inspections set according to project type. The abnormal period inspection ratio is the ratio of the number of inspections during nighttime or holiday periods to the total number of inspections. The irregularity of the inspection interval is the standard deviation of the time interval between two adjacent inspections in the past year.
6. The AI-based automatic doping risk identification system according to claim 1, characterized in that: The XGBoost classification model is trained using a cost-sensitive learning strategy, where the ratio of the weight of positive samples to the weight of negative samples is equal to the ratio of the number of negative samples to the number of positive samples in the training set.
7. An AI-based automatic doping risk identification method, applied to the AI-based automatic doping risk identification system as described in any one of claims 1-6, characterized in that: Includes the following steps: Step 1: Obtain athletes' basic information, whereabouts information, doping test results, and outcome management data; Step 2: Calculate the performance mutation index and coach violation mark based on the athlete's basic information data; calculate the ambiguity of whereabouts and the abnormality of whereabouts update frequency based on the whereabouts information data; calculate the inspection coverage rate, the proportion of inspections during abnormal periods, and the irregularity of inspection intervals based on the doping inspection data; calculate the number of past violations and the number of days from the end of the suspension to the current date based on the results management data; and then combine the performance mutation index, coach violation mark, whereabouts ambiguity, abnormality of whereabouts update frequency, inspection coverage rate, proportion of inspections during abnormal periods, irregularity of inspection intervals, number of past violations, and the number of days from the end of the suspension to the current date into a feature vector. Step 3: Input the feature vector into the pre-trained XGBoost classification model and output the risk probability value; Step 4: Compare the risk probability value with the preset risk threshold. When the risk probability value is greater than or equal to the preset risk threshold, output a risky label. When the risk probability value is less than the preset risk threshold, output a risk-free label.
8. The method for automatic identification of doping risk using AI according to claim 7, characterized in that: In step two, the time span of the tracking update frequency anomaly index is the past 365 days, which is the time interval sequence of the tracking report timestamps.
9. The method for automatic identification of doping risk using AI according to claim 7, characterized in that: In step three, the training data of the XGBoost classification model is divided into a training set, a validation set, and a test set in chronological order. The training set is used for learning model parameters, the validation set is used for hyperparameter tuning, and the test set is used for final performance evaluation.
10. The method for automatic identification of doping risk using AI according to claim 7, characterized in that: In step four, the preset risk threshold is 0.5.