A stimulant risk intelligent assessment system and method

Abstract

This invention discloses an intelligent doping risk assessment system and method. The system includes a data acquisition and processing module, a feature extraction module, an intelligent risk assessment module, and an output and decision-making module. The method includes steps one, data acquisition and preprocessing; step two, multi-dimensional feature extraction; step three, intelligent risk assessment; step four, risk level determination; and step five, result generation and output. This invention can automatically integrate and analyze athletes' behavioral patterns, historical records, and other risk characteristics, outputting quantified risk levels and clear decision-making criteria. This enables accurate identification and focused monitoring of high-risk targets, significantly improving the deterrent effect and work efficiency of inspections, optimizing resource allocation, reducing overall costs, and providing objective and interpretable data support for targeted education and management of athletes.

Description

Technical Field

[0001] This invention relates to the field of sports management technology, and in particular to an intelligent doping risk assessment system and method. Background Technology

[0002] Doping has always been a sensitive issue of public concern, and international doping incidents are often intertwined with political maneuvering and closely related to national image. This necessitates that we accurately assess doping risks, take targeted preventative measures, and avoid doping incidents from occurring.

[0003] Current anti-doping efforts generally employ a "one-size-fits-all" testing frequency, lacking specificity and resulting in inefficient allocation of testing resources. Furthermore, reliance on expert experience for risk prediction is highly subjective and difficult to implement on a large scale and in a standardized manner. Moreover, the lack of dynamic and comprehensive quantitative analysis of athletes' multi-dimensional risk factors makes it impossible to accurately and dynamically assess individual athletes' doping risks, hindering the concentration of limited testing resources on the highest-risk targets. Therefore, this invention proposes an intelligent doping risk assessment system and method to address the problems existing in the prior art. Summary of the Invention

[0004] To address the aforementioned problems, the present invention aims to propose an intelligent doping risk assessment system and method. This system and method comprehensively analyzes an athlete's basic information, whereabouts, number of tests conducted, testing institutions, and doping violations to achieve automated and intelligent assessment of an individual athlete's doping risk, thereby making anti-doping testing more precise and efficient.

[0005] To achieve the objectives of this invention, the following technical solution is provided: an intelligent doping risk assessment system, comprising a data acquisition and processing module, a feature extraction module, an intelligent risk assessment module, and an output and decision module. The data acquisition and processing module is used to collect and standardize raw athlete data from multiple data sources. The feature extraction module is used to extract multidimensional risk feature vectors from the standardized data. The intelligent risk assessment module calculates a comprehensive risk score based on the multidimensional risk feature vectors and a pre-trained gradient boosting decision tree model, and then outputs a risk level of high risk, medium risk, or low risk according to a preset threshold. The output and decision module is used to visually display the risk level, comprehensive risk score, and main risk basis.

[0006] Further improvements are made in that: the feature extraction module specifically includes an athlete basic information feature unit, a location information feature unit, an inspection status feature unit, and a result management feature unit. The athlete basic information feature unit generates a first feature vector based on the athlete's basic information data. The location information feature unit generates a second feature vector based on the athlete's movement location information. The inspection status feature unit generates a third feature vector based on the athlete's historical inspection data. The result management feature unit generates a fourth feature vector based on the athlete's historical management data.

[0007] Further improvements are made in that: the athlete's basic information data includes the sequence of competition results, the type of sport, dietary records, and the history of violations by associated coaches. The first feature vector is obtained by calculating the performance stability score, the sport risk base score, the dietary risk score, and the associated coach risk score. The athlete's whereabouts information includes a sequence of whereabouts locations, whereabouts update records, and information accuracy. A second feature vector is obtained by calculating a geographic location risk score, an update compliance score, and an information quality score.

[0008] Further improvements are made in that: the historical inspection data includes historical inspection frequency and type, inspection ratio during abnormal periods, application status of drug and therapeutic drug exemptions and reported records, and a third feature vector is obtained by calculating inspection density and intensity scores, abnormal period inspection ratio scores, drug risk scores and reported identifiers; The historical control data includes the types of historical doping violations, the results of penalties, and the time of the violations. A fourth feature vector is obtained by calculating the historical violation severity index and the time-based recidivism risk attenuation factor.

[0009] Further improvements are made in the following aspects: The intelligent risk assessment module includes a feature splicing unit, a risk assessment model unit, a risk level classification unit, and a feature contribution analysis unit. The feature splicing unit is used to splice the extracted multi-dimensional risk feature vectors to form a comprehensive feature vector. The risk assessment model unit is used to process the comprehensive feature vector through a trained gradient boosting decision tree model and output a comprehensive risk score. The risk level classification unit is used to compare the comprehensive risk score with a preset threshold and classify it into high risk, medium risk, and low risk. The feature contribution analysis unit is used to analyze the decision-making process of the gradient boosting decision tree model, identify and output the top three specific risk feature indicators and their contribution values ​​that contribute to the current risk level assessment, as the main risk basis.

[0010] An assessment method for a doping risk intelligent assessment system includes the following steps: Step 1: Collect multi-dimensional element data of the target athlete and perform cleaning and standardization processing; Step 2: Extract quantitative risk characteristics from the standardized data across four dimensions: athlete basic information, whereabouts information, doping test results, and outcome management. Step 3: Combine all the extracted quantitative risk features into a comprehensive feature vector, and input it into the pre-trained gradient boosting decision tree model to obtain the comprehensive risk score; Step 4: Compare the comprehensive risk score with the preset threshold to determine whether the target athlete's doping risk level is high, medium, or low and output the result. Step 5: Output the risk level, overall risk score, and the main risk basis that led to the assessment result.

[0011] Further improvements are made in the following aspects: Step 2 involves extracting risk characteristics of athletes' basic information, including calculating the coefficient of variation of the competition results sequence to obtain a performance stability score, assigning a risk base score to the event based on whether the event is physical fitness-related, calculating a dietary risk score based on the frequency of eating out, and assigning a risk score to the associated coach based on whether the associated coach has a history of doping violations.

[0012] Further improvements are made in the following aspects: Step 2 extracts risk features of location information, including calculating a geographic location risk score based on whether the location of the location belongs to a preset set of remote or high-risk areas, calculating an update compliance score based on the proportion of successfully updated location information within a specified time, and calculating an information quality score based on the proportion of fuzzy records of location information.

[0013] A further improvement is that the preset threshold in step four includes a high-risk threshold θ. high and low-risk threshold θ low When the overall risk score is ≥ θ high It is judged as high risk; when θ high >Comprehensive risk score ≥θ low The risk level is determined to be medium; when the overall risk score is less than θ. low It was determined to be low risk.

[0014] The beneficial effects of this invention are as follows: This invention can automatically integrate and analyze athletes' behavioral patterns, historical records and other risk characteristics, output quantitative risk levels and clear decision-making basis, thereby achieving accurate identification and key monitoring of high-risk targets. While significantly improving the deterrent effect and work efficiency of inspections, it optimizes resource allocation and reduces overall costs. It also provides objective and interpretable data support for targeted education and management of athletes. Attached Figure Description

[0015] Figure 1 This is a module architecture diagram of the evaluation system of the present invention.

[0016] Figure 2 This is a flowchart of the evaluation method of the present invention. Detailed Implementation

[0017] 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.

[0018] Example 1 according to Figure 1 As shown in the figure, this embodiment provides a doping risk intelligent assessment system, including a data acquisition and processing module, a feature extraction module, an intelligent risk assessment module, and an output and decision-making module.

[0019] The data acquisition and processing module is used to collect and standardize raw athlete data from multiple data sources.

[0020] Specifically, raw athlete data is collected from multiple specified heterogeneous data sources via network interfaces. The collected raw data is then cleaned, formatted, and structured to generate standardized athlete data records. The data sources include athlete registration and performance databases, location information reporting platforms, doping control databases, and official penalty announcement databases.

[0021] The feature extraction module is connected to the data acquisition and processing module to extract multi-dimensional risk feature vectors from standardized data. The vectors cover four dimensions: athlete basic information, whereabouts information, doping test results, and historical results. Specifically, they include athlete basic information feature units, whereabouts information feature units, test results feature units, and result management feature units. The athlete basic information feature unit generates a first feature vector based on the athlete's basic information data; the whereabouts information feature unit generates a second feature vector based on the athlete's movement tracking information; the inspection status feature unit generates a third feature vector based on the athlete's historical inspection data; and the result management feature unit generates a fourth feature vector based on the athlete's historical management data. Athlete basic information data includes competition performance series, type of sport, dietary records, and history of violations by associated coaches, which is used to calculate a performance stability score S. stability Project risk baseline S project Dietary risk score S diet and associated coach risk score S coach The first feature vector is obtained, and the score calculation specifically includes: Performance stability rating S stability Based on the athlete's historical competition results sequence P={p1,p2,…,p n} Calculate, the formula is S stability=1-min(CV,1), where CV is the coefficient of variation, CV=σ(p) / μ(p), where σ(p) is the standard deviation and μ(p) is the mean; Project risk baseline S project The athlete is assigned a value directly based on the type of sport they participate in, T, where T is a discrete variable. Physical fitness sports are assigned a value of 1, and skill-based sports are assigned a value of 0. Dietary Risk Score diet Based on the frequency of athletes dining out, F out Calculate, the formula is S diet =min(F out ,1), F out It is the ratio of the number of times one eats out per unit of time to the total number of times one eats out; Associated Coach Risk Score coach The value C is directly assigned based on whether the athlete's associated coach has a history of doping violations; if so, C=1, otherwise C=0.

[0022] Athlete location information includes location sequence, location update records, and information accuracy, which is calculated by a geographic location risk score S. location Update compliance score S compliance Information quality score S quality The second feature vector is obtained, and the score calculation includes: Geographical location risk score S location Based on the set of locations reported by athletes, L={loc1,loc2,…,loc... n} Calculate, the formula is S location =count(loc n ∈R set ) / |L|, where R set Let L be a predefined set of high-risk or remote geographical locations, where count(·) is the counting function and |L| is the total number of locations. Update compliance score S compliance The calculation is based on the update time record of the location information, and the formula is S. compliance =Non time / N total Where Non time represents the number of times an update was successfully performed within the specified time window, N total This represents the total number of updates required. Information quality score S quality The proportion A of fuzzy records based on location information is calculated using the formula S. quality =1-A, where A is the proportion of the number of tracking records marked as "fuzzy" or unclear out of the total number of reported records.

[0023] Historical inspection data includes the frequency and type of historical inspections, the proportion of inspections during abnormal periods, the status of applications for drug and therapeutic drug exemptions, and records of whistleblowing. This data is used to calculate an inspection density and intensity score (S). frequency , Ratio of inspections during abnormal periods (S) abnormalTime Drug Risk Score med The third feature vector is obtained from the reported identifier, and the score calculation includes: Check density and strength score S frequency Calculated based on historical inspection records, the formula is S. frequency =α N total +β N blood , where N total N represents the total number of inspections within a specific time period. blood The number of blood tests or dried blood spot examinations within the same time period, α and β are preset weights, and β>α; Abnormal period inspection ratio score S_ abnormalTime Calculate the proportion of inspections conducted during preset abnormal periods (such as late at night or extremely short notification times) out of the total number of inspections; Drug Risk Score med By comparing the athletes' declared drug list with the list of prohibited drugs, and taking into account the therapeutic use exemption status, the highest risk score is given if a prohibited drug without a valid exemption is used, a medium risk score is given if a specific drug requiring attention is used, and a low risk score is given otherwise.

[0024] Historical control data includes the types of historical doping violations, penalties, and the time of the violation. This data is used to calculate the historical violation severity index S. history and time-based recidivism risk decay factor S recency The fourth feature vector is obtained, and the score calculation includes: Historical Violation Severity Index S history Based on previous violation records, the calculation formula is S. history =Σ i (w i severity i ), where severity i To map the severity score based on the type of violation and the duration of the suspension for the i-th violation, w i w represents the weights that decay over time. i =γ exp(-λ Δt i ), Δt i γ and λ are preset coefficients representing the time interval from the current point in time. Recidivism time decay factor Srecency Based on the time Δt since the most recent violation, the calculation formula is S. recency =exp(-k Δt), where k is the preset attenuation coefficient.

[0025] The intelligent risk assessment module is connected to the feature extraction module. Based on multi-dimensional risk feature vectors, it calculates a comprehensive risk score using a pre-trained gradient boosting decision tree model, and then determines the risk score based on a preset threshold (high-risk threshold θ). high and low-risk threshold θ low It outputs risk levels of high, medium, or low risk, including feature splicing units, risk assessment model units, risk level classification units, and feature contribution analysis units.

[0026] The gradient boosting decision tree model is trained using a historical athlete dataset. Each sample's features are feature vectors X extracted by the multidimensional risk feature extraction module, and the label y is a binary variable (1 indicates that a doping violation occurred subsequently, and 0 indicates that it did not occur). The model is an additive model, and its final output function F(X) is composed of K decision trees (base learners), expressed as: F(X) = ΣK k=1v h k (X), where h k (X) represents the predicted output of the k-th decision tree, and v represents the learning rate; The training process is performed using a forward step-by-step algorithm, iteratively adding new decision trees h. k To minimize the specified loss function L(y,F(X)), the negative gradient of the current model residual is fitted in each iteration.

[0027] The feature concatenation unit is used to concatenate the extracted multidimensional risk feature vectors to form a comprehensive feature vector; The risk assessment model unit is used to process the comprehensive feature vector through a trained gradient boosting decision tree model and output a comprehensive risk score; The risk level classification unit is used to compare the comprehensive risk score with a preset threshold and classify it into high risk, medium risk and low risk; The feature contribution analysis unit is used to analyze the decision-making process of the gradient boosting decision tree model, identify and output the top three specific risk feature indicators and their contribution values ​​that contribute to the current risk level assessment, serving as the primary risk basis. Specifically: When using the gradient boosting decision tree model for prediction, SHAP or a tree-based feature contribution calculation method is used to quantify the contribution of each feature component in the input comprehensive feature vector to the final comprehensive risk score (or risk level). During output, the features are sorted from high to low based on their absolute contribution values, and the top three feature indicators and their specific contribution values ​​are selected as the "top N specific feature indicators with the highest contribution to the risk level" for output.

[0028] The output and decision-making module is connected to the intelligent risk assessment module to visualize the risk level, comprehensive risk score and main risk basis. The main risk basis is the top N specific characteristic indicators and their contribution values ​​that contribute the most to the risk level.

[0029] Example 2 according to Figure 2 As shown in the figure, this embodiment provides an assessment method for a doping risk intelligent assessment system, including the following steps: Step 1: Collect multi-dimensional element data of the target athlete and perform cleaning and standardization processing; Step 2: Call the various units of the feature extraction module to extract quantitative risk features from the standardized data in four dimensions: athlete basic information, whereabouts information, doping test results, and result management. The process of extracting risk characteristics from athletes' basic information includes calculating the coefficient of variation of the competition results sequence to obtain a performance stability score, assigning a risk base score to the event based on whether the event is physical fitness-related, calculating a dietary risk score based on the frequency of eating out, and assigning a risk score to the associated coaches based on whether the associated coaches have a history of doping violations. Extracting risk characteristics from location information includes calculating a geographic location risk score based on whether the location belongs to a pre-defined set of remote or high-risk areas, calculating an update compliance score based on the proportion of successfully updated location information within a specified time, and calculating an information quality score based on the proportion of fuzzy location information records.

[0030] Step 3: Combine all extracted quantitative risk features into a comprehensive feature vector V. combined The input is fed into a pre-trained gradient boosting decision tree model to obtain a comprehensive risk score, Risk. Score =F(V combined ).

[0031] Step 4: Calculate the overall risk score (Risk). Score With the preset threshold (θ) high and θ low The system compares the doping risk levels of the target athletes to determine whether they are high-risk, medium-risk, or low-risk, and outputs the results. When Risk Score ≥θhigh It is judged as high risk; when θ high >Risk Score ≥θ low The risk level is determined to be medium; when Risk Score <θ low It was determined to be low risk.

[0032] Step 5: Output the risk level, overall risk score, and the main risk basis for the assessment result; Specifically, the assessment report is generated, which lists the risk level and comprehensive risk score, and calls the feature contribution analysis unit to output the three feature indicators that have the greatest impact on the assessment results and their contribution.

[0033] 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 invention without departing from its spirit and scope, 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. A doping risk intelligent assessment system, characterized in that: The system includes a data acquisition and processing module, a feature extraction module, an intelligent risk assessment module, and an output and decision-making module. The data acquisition and processing module is used to collect and standardize raw athlete data from multiple data sources. The feature extraction module is used to extract multi-dimensional risk feature vectors from the standardized data. The intelligent risk assessment module calculates a comprehensive risk score based on the multi-dimensional risk feature vectors and a pre-trained gradient boosting decision tree model, and then outputs a risk level of high risk, medium risk, or low risk according to a preset threshold. The output and decision-making module is used to visually display the risk level, comprehensive risk score, and main risk basis.

2. The doping risk intelligent assessment system according to claim 1, characterized in that: The feature extraction module specifically includes an athlete basic information feature unit, a location information feature unit, an inspection status feature unit, and a result management feature unit. The athlete basic information feature unit generates a first feature vector based on the athlete's basic information data. The location information feature unit generates a second feature vector based on the athlete's movement tracking information. The inspection status feature unit generates a third feature vector based on the athlete's historical inspection data. The result management feature unit generates a fourth feature vector based on the athlete's historical management data.

3. The doping risk intelligent assessment system according to claim 2, characterized in that: The athlete's basic information data includes the sequence of competition results, the type of sport, dietary records, and the history of violations by associated coaches. The first feature vector is obtained by calculating the performance stability score, the sport risk base score, the dietary risk score, and the associated coach risk score. The athlete's whereabouts information includes a sequence of whereabouts locations, whereabouts update records, and information accuracy. A second feature vector is obtained by calculating a geographic location risk score, an update compliance score, and an information quality score.

4. The doping risk intelligent assessment system according to claim 2, characterized in that: The historical inspection data includes historical inspection frequency and type, inspection ratio during abnormal periods, application status for drug and therapeutic drug exemption, and reported records. A third feature vector is obtained by calculating inspection density and intensity scores, abnormal period inspection ratio scores, drug risk scores, and reported identifiers. The historical control data includes the types of historical doping violations, the results of penalties, and the time of the violations. A fourth feature vector is obtained by calculating the historical violation severity index and the time-based recidivism risk attenuation factor.

5. The doping risk intelligent assessment system according to claim 1, characterized in that: The intelligent risk assessment module includes a feature concatenation unit, a risk assessment model unit, a risk level classification unit, and a feature contribution analysis unit. The feature concatenation unit concatenates extracted multi-dimensional risk feature vectors to form a comprehensive feature vector. The risk assessment model unit processes the comprehensive feature vector using a trained gradient boosting decision tree model and outputs a comprehensive risk score. The risk level classification unit compares the comprehensive risk score with a preset threshold and classifies it into high-risk, medium-risk, and low-risk categories. The feature contribution analysis unit analyzes the decision-making process of the gradient boosting decision tree model, identifies and outputs the top three specific risk feature indicators and their contribution values ​​that contribute to the current risk level assessment, serving as the primary risk basis.

6. The assessment method of the intelligent doping risk assessment system according to any one of claims 1-5, characterized in that, Includes the following steps: Step 1: Collect multi-dimensional element data of the target athlete and perform cleaning and standardization processing; Step 2: Extract quantitative risk characteristics from the standardized data across four dimensions: athlete basic information, whereabouts information, doping test results, and outcome management. Step 3: Combine all the extracted quantitative risk features into a comprehensive feature vector, and input it into the pre-trained gradient boosting decision tree model to obtain the comprehensive risk score; Step 4: Compare the comprehensive risk score with the preset threshold to determine whether the target athlete's doping risk level is high, medium, or low and output the result. Step 5: Output the risk level, overall risk score, and the main risk basis that led to the assessment result.

7. The assessment method of the intelligent doping risk assessment system according to claim 6, characterized in that: Step two involves extracting risk characteristics of athletes' basic information, including calculating the coefficient of variation of the competition results sequence to obtain a performance stability score, assigning a risk base score to the event based on whether the event is physical fitness-related, calculating a dietary risk score based on the frequency of eating out, and assigning a risk score to the associated coach based on whether the associated coach has a history of doping violations.

8. The assessment method of the intelligent doping risk assessment system according to claim 6, characterized in that: Step two involves extracting risk features from location information, including calculating a geographic location risk score based on whether the location belongs to a preset set of remote or high-risk areas, calculating an update compliance score based on the proportion of successfully updated location information within a specified time, and calculating an information quality score based on the proportion of fuzzy location information records.

9. The assessment method of the intelligent doping risk assessment system according to claim 6, characterized in that: The preset thresholds in step four include a high-risk threshold θ. high and low-risk threshold θ low When the overall risk score is ≥ θ high It is judged as high risk; when θ high >Comprehensive risk score ≥θ low The risk level is determined to be medium; when the overall risk score is less than θ. low It was determined to be low risk.

Images