Computer-implemented method for carrying out a performance analysis of a user

Abstract

The invention relates to a computer-implemented method for carrying out a performance analysis of a first user (B1), the method comprising the steps of: - providing a database (1) having a plurality of data records (DSi), wherein each data record (DSi) comprises a plurality of items of personal data (pD) and at least one first performance measurement value (LM); - receiving items of personal data (pDB1-1, pDB1-2) of the first user (B1) and a measured performance value (LMB1-1) of the first user (B1); - determining at least two different subgroups (UG1, UG2, UG3) of the database (1); - determining, for said subgroups (UG1, UG2, UG3), at least one statistical property; - using that subgroup (UG1, UG2, UG3) which, according to the at least one determined statistical property, is most representative of the first user (B1); and - carrying out the performance analysis of the first user (B1) by determining a performance result of the performance measurement value (LMB1-1) of the first user (B1) within the subgroup (UG1, UG2, UG3) that is used.

Description

[0001] Computer-implemented method for conducting a user performance analysis

[0002] The invention described herein relates to a computer-implemented method for performing a performance analysis of a user.

[0003] Performance analysis plays a central role in numerous fields, particularly in rehabilitation, physiotherapy, and competitive sports. In physiotherapy, objective performance assessment allows therapists to evaluate a patient's condition, track progress, tailor therapy plans, and ensure evidence-based rehabilitation. For competitive athletes, continuously monitoring their physical condition and training progress is crucial to maximizing the effectiveness of their training program and minimizing the risk of injury. Therefore, the availability of accurate and efficient analytical methods is essential for making informed decisions regarding training or therapy.

[0004] Existing performance analyses are traditionally conducted using manual methods or simple electronic systems. In competitive sports, measuring devices such as heart rate monitors, pedometers, or force sensors are frequently used to record specific parameters like heart rate, cadence, or force development. However, such devices often provide only isolated data, making a holistic analysis difficult. Furthermore, they often require manual interpretation of the results by coaches or athletes, which can be time-consuming and prone to error.

[0005] Another problem is that therapists need to be familiar with a wide variety of measurement systems in order to interpret their results meaningfully. This requires not only comprehensive expertise but also considerable experience, especially when the data is considered in the context of a specific condition. Furthermore, it has been shown that therapists and trainers are not always free from bias when analyzing data. This can lead to performance analyses being subjectively influenced and therefore not always reproducible.

[0006] In physiotherapy and rehabilitation, digital solutions such as wearable sensors or apps that record movement patterns and biomechanical parameters are increasingly being used. However, these systems are often limited to specific exercises or devices and offer only limited possibilities for integrating different data sources. Furthermore, a solution that allows the collected information to be placed in a meaningful context and automatically evaluated is often lacking.

[0007] The present invention aims to overcome the aforementioned disadvantages and to provide a method that enables flexible, precise and data protection-compliant performance analysis.

[0008] This goal is achieved through a computer-implemented procedure for conducting a performance analysis of a first user, comprising the following steps:

[0009] - Providing a database containing a large number of data records, each data record comprising a large number of personal data (which may be anonymized) and at least one initial performance metric,

[0010] - Receiving personal data of the first user and a measured performance value of the first user or an indication of a performance value to be measured of the first user,

[0011] - Determining at least two distinct subsets of the database, each subset comprising only those records of the database where a selected date or combination of at least two personal data items matches or is related to a corresponding date or combination of personal data of the first user,

[0012] For the aforementioned subgroups, at least one statistical property, for example a variance or standard deviation, must be determined from the performance measurements of the subgroups.

[0013] - Selecting the subgroup that is most representative of the first user according to at least one determined statistical property, preferably by performing a discriminant analysis,

[0014] - Conducting the performance analysis of the first user by determining a performance result of the first user's performance measurement within the subgroup used.

[0015] The method according to the invention enables the classification of the user's performance results within the context of their own performance potential. Specifically, the user's performance result can be displayed within the context of the group of people that is most representative of them, and not in an arbitrarily chosen group. This allows, for example, targeted adjustments to training plans, particularly in physiotherapy, rehabilitation, and for competitive athletes. In general, however, the method according to the invention enables an objective assignment of the measured performance value to the specific user, which is a highly complex task due to the large amount of available personal data. The invention particularly prevents subjective misjudgments by doctors, physiotherapists, trainers, etc., which can, for example, lead to...Previously, users had to assess, based on experience and therefore subjectively, whether a performance measurement XY was good or bad for a patient aged 65 to 75 years with above-average weight and a specific medical history (this is an arbitrary and non-limiting example). In summary, the output of the procedure is a relative, but still objective, performance result for the user. The objectivity of the relative performance result is ensured by the statistical analysis of the subgroups.

[0016] For example, a patient might have suffered a stroke, resulting in limited hand mobility. Physiotherapy can improve hand mobility, and current technology allows for the measurement of gradual progress (e.g., a weekly increase in performance) – this can be described as a relative context. However, the inventive method places the measured performance values ​​in an absolute context with respect to the most relevant group of people for the user. Thus, it does not establish an arbitrary group reference, whereby the individual's performance is placed in an absolute context compared to the group "gender: male," in which the stroke patient would most likely be ranked at the lower end of the performance spectrum.According to the invention, a statistical method is used to determine the most relevant group of people, which in this case could be, for example, the group "disease: stroke". The statistical method could also, for example, determine as the most relevant group of people the group formed by the combination of the personal data "gender: male" and "disease: stroke". Within the context of these groups, the patient can view their performance measurements in an absolute context, so that, for example, they achieve a performance result in the upper third of this performance group, whereas in the "male" group they might only expect a performance result in the lower fifth. It should be noted that the more data is available, the more specific the group classification can be.

[0017] For example, the combination of data points "Disease: Stroke," "Sex: Male," "Age group: 30-40," and "Weight class: Obesity 1" could be identified as the most relevant subgroup if sufficient data is available. As mentioned previously, classifying performance results in absolute context can help to tailor training, physiotherapy, and rehabilitation plans more effectively. In particular, the absolute context of the most relevant subgroup clearly reveals whether a disease is severe, thus requiring, for example, a gentler approach to physiotherapy or rehabilitation, or whether the disease is less severe, allowing for more advanced physiotherapy or rehabilitation.

[0018] Furthermore, it should be noted that the invention also makes it possible, for the first time, to discover novel relationships between performance and personal data that were previously unknown in the professional community (e.g., to the physiotherapist or coach of the competitive athlete). For example, the method according to the invention may reveal that some performance measurements are strongly dependent on the combination of certain personal data that were previously unknown.

[0019] Thus, the statistical method can reveal particularly "surprising" representative subgroups for the user, which would not have been readily apparent even to a trained physiotherapist or trainer. For example, a performance metric might be significantly influenced by the combination of two personal data points, whereas this same combination might have little impact on other performance metrics. Consequently, the method might reveal that for a female individual with a stroke and Parkinson's disease, the most relevant subgroup for a specific performance metric is the one with the combination of the personal data "gender: female" and "disease: Parkinson's disease," and not the one with the personal data "disease: stroke" (fictitious example).It is therefore particularly advantageous if at least one of the subgroups for which a statistical property is determined includes only those records of the database where a selected combination of at least two personal data items matches or is related to a corresponding combination of the personal data of the first user.

[0020] At this point, the term "related" should also be explained. Personal data records might, for example, include a precise date of birth. For the purpose of correlation ("related"), however, certain age groups can be assumed, e.g., under 30 years, 30 to 50 years, and over 50 years. Within these three age groups, two dates of birth, each indicating an age over 50, would be related. Generally, two pieces of personal data can be defined as related if they can be grouped into a common, predefined category.

[0021] Furthermore, the feature "receiving a measured performance value from the first user or an indication of a performance value to be measured from the first user" is also explained. For the step of determining at least one subgroup, no performance value needs to be known at all initially. For the step of determining at least one statistical property, in the simplest case, only an indication can be given as to which performance values ​​the statistical property should be determined. The actually measured performance result is only considered optionally here. In the step of performing the performance analysis, the actually measured performance value is used. To measure the performance value, a user can perform the following step: performing a physical or mental exercise and measuring the performance value.The mental exercise could be, for example, a math problem or something similar.

[0022] The present invention is particularly applicable in the context of so-called "serious games" in medicine and therapy, e.g., in rehabilitation games, such as those for stroke patients to improve hand motor skills. The method according to the invention allows the difficulty level of these "serious games" to be adjusted more precisely and provides the user with better feedback on their performance.

[0023] The determined performance result can be displayed on a screen and / or fed into subsequent data processing. Furthermore, it should be noted that the computer-implemented method can be carried out on a single computer or on a distributed computing system. The computer or computing system can each have one or more processors for performing the necessary calculations and may have one or more interfaces through which data can be input and / or output. The computer or computing system can be connected to or include the database. In particular, the computer or computing system can...The computing system must be connected to the aforementioned screen for displaying the performance result and / or have an interface for receiving the performance measurements and / or the personal data of the first user (and / or a second user), whereby this interface may be an HMI (Human Machine Interface) through which the measured performance measurements or personal data can be manually entered, or may be an interface to a sensor that directly measures the performance measurement.

[0024] In a particularly preferred embodiment, all possible subgroups are determined in the step of identifying at least two distinct subgroups. This has the advantage that no subjective "pre-selection" can occur, which would be the case, for example, if a person were to subjectively (and therefore potentially incorrectly) determine beforehand that a gender-specific distinction will have no influence on the most relevant subgroup.

[0025] In another preferred approach, a statistical property is determined for a previously defined subgroup only if the subgroup is statistically relevant, particularly if it comprises a predetermined number of records. In other words, a subgroup is first created and then checked for statistical relevance. If a subgroup contains only a few entries, it can generally be assumed that it will not be statistically relevant, and therefore this subgroup is excluded from further processing. This step is particularly relevant if all possible subgroups have been determined beforehand, as some combinations of personal data occur only rarely and will therefore not be statistically relevant. It goes without saying that statistical relevance can also be determined in other ways.In general, statistical relevance describes the probability that an observed effect did not occur by chance but actually has a systematic cause. It can therefore typically be determined by significance tests, which examine whether an effect (e.g., a difference between two groups or a relationship between variables) is highly likely to exist in the population and is not merely due to random fluctuations in the sample. In the present context, statistical relevance can be determined, for example, via the sample size and / or the effect size. The sample size influences the power of the analysis, as larger samples reduce random fluctuations and make smaller effects detectable.The effect size indicates the strength of an observed effect, where small effects may be practically irrelevant despite significance, while large effects can be detected even with smaller samples.

[0026] Special measures can be taken when two different performance metrics are available for the same user. For example, a user has the personal data "illness: stroke" and "age: over 50" and begins a performance assessment, such as playing a marble run. The performance metrics could be, for example, "time required" and "distance traveled by the marble." For the first performance metric, "time required," the subgroups "illness: stroke" and "age: over 50" are analyzed, and it is determined that the most relevant subgroup is the one with the date "illness: stroke."Furthermore, for the second performance measurement, "distance traveled by the ball," both of the aforementioned subgroups are analyzed, and it is found that the most relevant subgroup is that with "Age: over 50," since, for example, a stroke has no effect on the distance traveled, and age is more relevant to the distance traveled (fictitious example). It is evident that the method according to the invention thus also enables an individual analysis of different performance measurements. Preferably, the method therefore comprises the following steps:

[0027] - Receiving a measured second power reading from the first user or receiving an indication of a second power reading to be measured from the first user,

[0028] Repeat the steps of identifying, gathering, and performing the second performance measurement.

[0029] In the aforementioned variant, it is preferred if different subgroups are used for the two performance metrics, with the subgroups preferably overlapping. In the example given with the subgroups "Disease: Stroke" and "Age: over 50," this can be the case, for example, since the subgroup "Disease: Stroke" may also include individuals who meet the criterion "Age: over 50."

[0030] In a further preferred embodiment, an additional subgroup can be used, and a further performance analysis of the first user can be performed by determining another performance result of the performance measurement within this additional subgroup. This additional subgroup need not be the most relevant subgroup, but could, for example, be a subgroup selected by the user. In particular, in this embodiment, both performance results can be displayed as a comparative representation on a screen. This allows the user to see how their performance result compares to all individuals who have had a stroke (most relevant group) and how their performance result compares to all male individuals (additional subgroup selected by the user). The additional subgroup could, for example, also represent a training goal, so that the user knows how to improve their performance to reach a target level.Preferably, the most relevant subgroup contains the information that a specific disease is present, and the further subgroup contains the information that the specific disease is not present.

[0031] Particularly advantageous is the use of the inventive procedure to compare performance results of two users who inherently have different performance maxima, for example, because one of the users has a specific illness or is particularly young or old. To implement this, the procedure can repeat the steps of receiving, determining, ascertaining, eliciting, and performing for a second user, where the type of performance measure of the first user corresponds to the type of performance measure of the second user (i.e., the two users performed the same exercise and the same performance measure was recorded, e.g., "time taken"), and where the elicited subgroups are different for the first and second users. In one example, a first user might have suffered a stroke, so the relevant subgroup for them would be the one with the personal data "illness: stroke".The second user could be a teenager, so the relevant subgroup for them is the one with the personal data "Age: under 30 years". Both users then play, for example, a marble run, where the performance metric is the "time taken". The users are then shown their own performance result in relation to their most relevant subgroup and can thus compare their two measurements. For example, the first user took longer to complete the marble run, but their performance result in their most relevant subgroup is 20% above average, while the second user's performance result in their most relevant subgroup is 10% above average. In this case, the first user would win the comparison despite taking longer.This method is particularly advantageous for patients undergoing therapy or rehabilitation, as the direct comparison with others is highly motivating and can thus accelerate the therapy or rehabilitation process. Since a patient improves through recovery, it can be assumed that their performance results will steadily improve. This may lead to a different subgroup being identified as the most relevant at a later stage. This allows for a consistent level of difficulty, which can provide additional motivation.

[0032] In the aforementioned method, it is particularly preferred if the performance measurements of the first and second users are determined essentially simultaneously and the performance results of the first and second users are preferably displayed as a comparative representation on one screen. This allows for direct competition between the two users on-site, which is particularly beneficial and motivating for "serious games".

[0033] Particularly useful is the creation and display of an overarching ranking (a so-called equality ranking) based on the relative performance results of the first and second users. In this ranking, users are ordered not by their absolute performance measurements, but by their calculated relative performance results. For example, a young girl might achieve a performance result that places her in the top 5% of her age and gender group, while her father, with a better absolute performance measurement, only reaches the top quartile of his own age and gender group. In the equality ranking, the daughter would thus be placed ahead of her father, as her relative performance is more outstanding in her context. This ranking normalizes performance and creates a fair basis for comparison between individuals with fundamentally different physical or experience-related prerequisites.

[0034] The "Equality Ranking" can also be used as a technical control parameter for subsequent automated processes, particularly in the field of competitive multiplayer applications such as serious games or e-sports. In such an embodiment, the "Equality Ranking" 4 It is used as input for a matchmaking algorithm. The algorithm automatically creates fair matches by matching users with a similar rank in the Equality Ranking, rather than users with similar absolute performance scores. 4This brings together different players. In this variant, users are located in different places, and their respective computing units are connected, for example, via the internet. This ensures that, for instance, a talented beginner can compete against an experienced veteran if both demonstrate a comparably outstanding performance within their respective peer groups. This optimizes the user experience and motivation, which is a key technical challenge in game design.

[0035] It goes without saying that the aforementioned database should contain as many data records as possible. To continuously expand the database, it is planned that personal data and measured performance metrics will be added after the performance analysis has been carried out. However, criteria can also be established to include only selected data records in the database.

[0036] This allows, for example, measurements that are based on obvious measurement errors or deliberately falsified performance data to be filtered out in order to avoid distorting the database.

[0037] In a first approach, preference is given if at least one of the determined statistical properties is a variance or standard deviation of the performance measurements of the subgroups, and a subgroup is considered more representative the lower its variance or standard deviation. If the variance or standard deviation is the only determined statistical property, the subgroup with the lowest variance or standard deviation is selected as the most representative subgroup.

[0038] In a second variant, at least one of the determined statistical properties can depend on the performance measurement of the first user. This statistical property is preferably the distance between the first user's performance measurement and the mean of the subgroup's performance measurements. A subgroup is more representative the smaller the distance between the first user's performance measurement and the mean of the subgroup's performance measurements. If the distance is the only statistical property, the subgroup with the smallest distance is selected as the most representative subgroup.

[0039] Preferably, at least two statistical properties are determined, and the most representative subgroup is identified as a function of all determined statistical properties. It should be noted, however, that this procedure can also be performed using a machine learning algorithm, and an analytical determination of the most representative subgroup is not always necessary.

[0040] In a particularly advantageous embodiment, the performance result of the first user is used to automatically create or adjust a training or therapy plan. The plan, which might be a digital sequence of exercises with specific duration, intensity, or number of repetitions, is modified or created by a computer system without human intervention. Selecting the most representative subgroup is crucial, as it ensures that the automatic adjustment is based on a statistically valid and user-relevant comparison. For example, the system can automatically suggest an easier exercise variation if the user's performance is at the lower end of their most representative group, or increase the intensity if the user demonstrates above-average performance.This avoids subjective misjudgments and enables objective, reproducible, and continuous adaptation of the therapy. Furthermore, the performance result determined using the method according to the invention can be used for computer games, in particular computer-based rehabilitation games, so-called 'serious games'. In a first variant, the performance measurement can be used for the automatic adjustment of at least one parameter of a computer game, in particular for adjusting the difficulty level. For example, the system can dynamically adjust the reaction time of computer-controlled opponents, the complexity of a task (e.g., a ball maze mentioned in claim 8), or the force or speed required to perform an action in the game. The result of the performance analysis is thus directly converted into a control parameter for the game engine.This creates an adaptive environment, particularly a therapeutic environment, that continuously challenges the user without over- or under-challenging them, thus improving performance or therapy outcomes.

[0041] Furthermore, the performance result can be provided as input to a matchmaking algorithm, which then assigns at least two players based on this result. This allows, for example, two individuals with widely differing absolute scores, but who objectively exhibit roughly equivalent relative performance results based on their personal data, to be matched. Since the two players are each in the top 10% of their respective most representative subgroups, these players, who relatively achieve the best results within their available resources, can be paired together.

[0042] Instead of, or in addition to, a purely visual display of the performance result, it can also be used to directly control a technical device that provides the user with immediate, physical feedback. The performance result is converted into a control signal for one or more actuators. For example, haptic feedback can be provided via a vibration motor in a handle or controller, the intensity or frequency of which reflects the user's performance relative to the average of their most representative subgroup. Alternatively, visual feedback can be provided via an LED strip, whose color or illumination duration indicates progress, or acoustic feedback, in which the pitch or rhythm of a signal changes with performance. In these cases, the analysis result is used directly to generate a physical effect.To ensure a high degree of objectivity and reproducibility in performance analysis, it is preferred that at least one performance measurement represents a value objectively measured by a sensor. Such values ​​are, in particular, biomechanical quantities such as a force exerted on a force sensor, a movement velocity recorded by accelerometer or gyroscope sensors, or a distance traveled determined by position tracking. Physiological quantities such as heart rate recorded by a pulse monitor can also serve as values. The use of such sensor-based, physical data ensures that the entire process, from data acquisition to the technical application of the result, is based on a real-world, measurable foundation and not on abstract or subjective assessments.Furthermore, such measured values ​​can be exercise-specific or game-specific measured values, such as, in particular, the time required to carry out an exercise or a game, a score achieved, or a characteristic value representing the quality of the exercise or game (such as the distance traveled by a ball in a ball labyrinth).

[0043] The aforementioned system can, for example, be designed using at least one computing unit or, more generally, a data processing device that is designed to carry out the aforementioned procedure.

[0044] The computer-implemented method described herein can be executed, in particular, by a performance analysis system. This system comprises, for example, at least one server or computing device containing at least one hardware processor, non-volatile memory, and a network interface. The memory contains computer-executable instructions which, when executed by the processor, configure the system to act as a specialized performance analysis unit. This unit is configured to receive personal data and performance metrics and to perform the steps described herein of determining subgroups, calculating statistical properties, and selecting the most representative subgroup. The system can be located on a dedicated physical server, a distributed cloud computing platform (e.g.,The performance analysis unit can be implemented as an Infrastructure as a Service (IaaS) or embedded within a larger therapeutic or training device. While the central processing may utilize conventional computing components, the system as a whole is distinguished by its integration with specific, non-generic input and output hardware to solve a technical problem inherent in measuring and adjusting physical performance. The performance analysis unit can be communicatively coupled with a variety of sensor devices designed to capture objective performance metrics from a user's physical activity. These sensor devices are typically specialized in measuring biomechanical, physiological, exercise-specific, or game-specific metrics.

[0045] Examples of such sensor devices include, but are not limited to: 1) Motion and orientation sensors, such as hardware accelerometers, gyroscopes, and inertial measurement units (IMUs), embedded in wearable devices or attached to the user's body to measure movement speed, range of motion, and technique. 2) Force and pressure sensors, such as force plates, dynamometers, or pressure-sensitive mats, designed to measure the physical force exerted by the user during exercise. 3) Physiological monitoring devices, such as heart rate monitors (e.g., electrocardiography sensors, ECG) or electromyography (EMG) sensors, to measure the user's physiological response to an activity.4) Position tracking systems, such as optical motion detection systems that use specialized cameras and markers, or magnetic tracking systems, to precisely track the user's body or an object in three-dimensional space. The data from these specific hardware sensors provide the objective, real-world basis for performance analysis, thus anchoring the method in a physical and technological context.

[0046] Furthermore, it may be preferable that the performance measurements are calculated from at least two performance measurement data points, which are preferably weighted in the performance measurement. A first example of this is the calculation of the total weight from the individual weights of three exercises in powerlifting. Here, the individual weights are not decisive, but rather the total weight calculated from them, since the overall performance analysis of an athlete is to be determined. In this example, the data sets could contain either the total weight or the individual weights, which could be converted into the total weight in a preliminary step. In other applications, however, the performance measurement data could also be converted into the performance measurement using weights. In one example, the two values ​​could have different orders of magnitude, for example, a maximum running distance and a maximum jump distance, and one could, for example,Before adding the two values, normalize them, for example, by scaling them to 0-100 between the lowest and highest expected values, although other weightings are also possible. In a second example, a first factor might be more relevant than a second. Here, you could weight both factors by multiplying each by a weighting factor, e.g., first factor *0.6 and second factor *0.4. Both approaches assume that there are multiple factors necessary for an overall performance evaluation and ranking. In yet other variations, performance measurement data with different units could be converted to performance measurement data, e.g., in boxing, a punch velocity in m / s and an impulse in N*s, if appropriate weighting is applied.It is evident from all the examples that the performance measurement data (and generally preferably also the performance values) are physical measurement parameters, preferably measurable or measured in SI units. In other examples, the performance measurement data or performance values ​​could also be chosen differently, e.g., as a number of points awarded in a sports exercise.

[0047] The output from the performance analysis unit can be displayed on a screen, for example. Furthermore, the output from the performance analysis unit can be used as a direct control signal for the automatic adjustment of the operation of a specific therapeutic or interactive output device and / or a computer game.

[0048] Preferred and non-restrictive embodiments are explained below by reference to the attached drawings.

[0049] Figure 1 shows a database with a large number of records and a user.

[0050] Figure 2 shows a first subset of the database.

[0051] Figure 3 shows a second subset of the database.

[0052] Figure 4 shows a third subset of the database.

[0053] Figure 5 shows a distribution of power measurements within the first subset. Figure 6 shows a distribution of power measurements within the second subset.

[0054] Figure 7 shows a distribution of performance measurements within the third subset of data. Figure 8 shows specific representation factors for the subgroups.

[0055] According to the present procedure, a database 1 is initially provided, in which a large number of data records DS1, DS2, ... DSi are stored. Database 1 can comprise hard drives, SSDs, or RAM. Database 1 can be implemented as a dedicated server or in the cloud.

[0056] In the example shown, each DSi data record comprises two personal data points (pD) and two performance metrics (LM). Personal data (pD) is any information relating to a natural person. Generally, each DSi data record can contain at least one or at least two personal data points (pD) and at least one or at least two performance metrics (LM).

[0057] Personal data (PD) can be anonymized and therefore cannot be linked to a specific individual. Personal data (PD) can include, for example, a date of birth or age, health data (e.g., weight, diagnoses, medical findings), genetic data, or similar information. Health data can, in particular, include information about a disease (e.g., the statement "disease X is present" or "disease Y is not present"). Generally, it is preferred that each data record (DSi) includes as much personal data (PD) as possible.

[0058] Performance measures (PMs) are measurement data representing the results of specific exercises. For example, a performance measure could be the number of successfully completed push-ups and / or squats. PMs can also be measured using tools, such as a ball maze, where the "time taken" and "distance traveled by the ball" are measured. These two examples illustrate that the PMs in a dataset (DSi) for an individual can be derived from various exercises, or they can be derived from the same exercise. A mix of performance measures is possible and generally preferred, with each dataset (DSi) containing as many PMs as possible.

[0059] The procedure described herein aims to evaluate the performance measurements LMB1 of a first user Bl using the data records DSi stored in database 1, so that these measurements can be output as performance results in relation to other individuals. Initially, it is sufficient to obtain the personal data pdB 1 of the first user Bl, and only then the actually measured performance measurements LMB1 or an indication of which performance measurements LMB1 should be measured. In the present example, two personal data points, pDBl-1 and pdBl-2, of the first user Bl are obtained, where pDBl-1 is the indication that an illness is present, and pdBl-2 is the weight of the first user Bl.

[0060] Using the personal data pDBl-1 and pDBl-2 of the first user Bl, those subgroups of data records DSi that are particularly representative of the first user Bl are to be extracted from the database DB. For this purpose, the data records DSi in the database DB are first divided into several subgroups.

[0061] Figures 2, 3, and 4 show subgroups created from the DSi records of database 1 for the first user Bl. For this example, it should be noted that the first user Bl in Figure 1 has provided the personal data "Disease X: yes" and "Weight: 75 kg".

[0062] The subgroups are determined by selecting a date from the personal data of the first user Bl, e.g., "Disease X: yes". Then, all records DSi from the database that also contain the specified date are searched, i.e., in the example given, "Disease X: yes". For database 1 shown in Figure 1, this would include at least records DSI, DS2, and DS3, but not records DS4 and DSi. These records DSI, DS2, and DS3 form a first subgroup UG1, as it only includes those records DSi in database 1 where the selected date "Disease X: yes" matches the corresponding date "Disease X: yes" in the personal data pDBl of the first user Bl.

[0063] A second subgroup can be formed by using a different piece of personal data from the first user Bl, e.g., "Weight: 75 kg". It is evident that no other weight in database 1 is exactly "75 kg", and for the evaluation, it is more relevant whether the weight is similar. For this reason, these personal data points should only be related. To this end, discrete value ranges are initially assumed as a temporary measure, and it is checked which records DSi in database 1 contain personal data that lies within the same discrete value range as the corresponding piece of personal data pDBl of the first user Bl, i.e., which records are related to this piece of personal data pDBl of the first user Bl.This is represented in Figure 3 by the discrete value range “60-80 kg”, so that the corresponding personal data of the records DSI and DS4 of database 1 of Figure 1 is in the same value range as the personal data of the first user B 1.

[0064] Furthermore, corresponding subgroups can be found, where the subgroup comprises only those records DSi from database 1 in which a selected combination of at least two personal data items matches or correlates with a corresponding combination of the personal data of the first user. In the case of Figure 1, a combination of the personal data "Disease X: yes" and "Weight: 75 kg" of the first user Bl is used. Database 1 is searched for records DSi that contain both "Disease X: yes" and "Weight: 60-80 kg" (as explained above, data can also be correlated here using a discrete range of values). In the example of Figure 1, this is only record DSI, as detailed in Figure 4.

[0065] Using this method, three subgroups UG1, UG2, and UG3 were determined from database 1, each representative of user 1. It goes without saying that many more subgroups can be created, for example, if more than two personal data points (pDBl) of the first user (Bl) exist, thus allowing for the determination of more subgroups with individually matching or correlating personal data points, or more subgroups with combinations of matching or correlating personal data points. It is also understood that subgroups with combinations of more than two matching or correlating personal data points can be formed.

[0066] As a rule, all possible subgroups are determined, although this is not always necessary, for example, if it is obvious that a subgroup makes no statistical sense, e.g., because it is too narrowly distributed and therefore will contain too few data records (DSi) in the subgroup, or too broadly distributed and therefore no statistically relevant statement can be made. For subgroups, it can therefore be checked beforehand whether they are statistically relevant, e.g., whether the subgroup comprises a predetermined number of data records or is too generic.

[0067] In the next step of the process, a statistically relevant property is determined for the subgroups in order to calculate a performance result for the first performance measurement LMB1-1 (the second performance measurement LMB1-2 of the first user is disregarded for the time being for the sake of simplicity). This is explained using Figures 5, 6, and 7.

[0068] Figures 5, 6, and 7 each show a distribution of the first power measurements LM (which are of the same type as the first power measurement LMB1-1 of the first user Bl under investigation) from the data records DSi of subgroups UG1, UG2, and UG3, respectively, as listed in database 1. The vertical axis represents the number N of data records DSi, and the horizontal axis represents the respective power measurement. It was assumed that database 1 contains significantly more data records DSi than shown in Figure 1, and the distribution curves 2, 3, and 4 were drawn over the discrete values. Thus, distribution curve 2 represents the distribution of the first power measurements of the first subgroup UG1, distribution curve 3 represents the distribution of the first power measurements of the second subgroup UG2, and distribution curve 3 represents the distribution of the first power measurements of the third subgroup UG3.

[0069] In a first example, the statistical property of the subgroup is determined as the variance or standard deviation (dispersion) of the respective subgroup. The variance is a statistical measure that indicates how much the values ​​of a data set spread around their mean M. It measures, therefore, how far the individual data points are, on average, from the mean M. The standard deviation is the square root of the variance. Comparing distribution curves 2, 3, and 4, it is evident that distribution curve 3 exhibits the greatest dispersion. It can therefore be concluded that subgroup UG2 is not considered the most representative subgroup in this example. This leaves subgroups UG1 and UG3, with subgroup UG3 exhibiting a slightly smaller dispersion than the first subgroup UG1 and is therefore considered the most representative subgroup in this example.

[0070] As background for the use of dispersion as a statistical property, it can be mentioned that, for example, in the case of a disease, similar symptoms may be present that will restrict the performance of all users affected by the disease in the same way, e.g., in the case of partial paralysis of a hand.

[0071] Figure 8 shows that, for this first example, a representation factor can be determined for each of the subgroups UG1, UG2, and UG3 with respect to a performance measurement LM or LMB1-1. In the first example, the representation factor is, for example, the inverse of the variance. The most representative subgroup is the one with the largest representation factor, i.e., in the first example, the third subgroup UG3.

[0072] The first example shows that the statistical property is determined without considering the first performance measurement LMB1-1 of the first user Bl. However, a statistical property can also be determined taking into account the first performance measurement LMB1-1 of the first user Bl. In particular, in a second example, the statistical property could be the distance between the first performance measurement LMB1-1 and the mean M in the respective subgroup.

[0073] Therefore, a large difference between the first performance measurement LMB1-1 and the mean value M indicates that it is unlikely that user 1 belongs to this subgroup, as can be seen particularly in Figure 7. If only this statistical property is considered, the second subgroup UG2 would be the most representative subgroup for the first user Bl.

[0074] The rationale for using the difference between the first performance measurement LMB1-1 and the mean value M is that, for example, symptoms of illnesses can be overcome through therapy or rehabilitation, meaning that a recovered user with a history of stroke can achieve a level of performance that is no longer comparable to that of typical stroke patients. Therefore, a recovered user with a history of stroke can once again be categorized within a broader group of people in terms of their performance capacity.

[0075] Figure 9 shows that, for this second example, a representation factor can be determined for each of the subgroups UG1, UG2, and UG3 with respect to a performance measurement LM or LMB1-1. In the second example, the representation factor is, for example, the reciprocal of the distance between the first performance measurement LMB1-1 and the mean M. The most representative subgroup is the one with the largest representation factor, i.e., in the second example, the second subgroup UG2.

[0076] To consider both aspects, at least two statistical properties of the subgroups are determined. Optionally, a first statistical property is determined based on the first performance measurement LMB1-1 of the first user Bl, and the second statistical property is determined without considering the first performance measurement LMB1-1 of the first user Bl. For example, both the variance and the distance of the first performance measurement LMB1-1 from the mean M can be determined. The most representative subgroup can be determined based on all determined statistical properties. By weighing the above findings, the first subgroup UG1 can be identified as the most representative subgroup, since, on the one hand, the variance is not too large, and on the other hand, the distance from the mean also indicates a plausible result.

[0077] Figure 10 shows that, for this third example, a representation factor can be determined for each of the subgroups UG1, UG2, and UG3 with respect to a performance measurement LM or LMB1-1. In the third example, the representation factor is, for instance, the inverse of the variance (Figure 8) multiplied by the inverse of the distance of the first performance measurement LMB1-1 from the mean M (Figure 9). It is understood, however, that the representation factor can be calculated using other mathematical models. The most representative subgroup is the one with the largest representation factor, i.e., in the third example, the first subgroup UG1.

[0078] It should be noted that the determined representation factors can also be used to draw conclusions about important performance factors. For example, using the calculation method shown in Figure 8, illness would have a 60% influence on the performance measurement and weight a 20% influence. It may therefore be preferable to output or display the representation factors for at least one, at least two, or all subgroups, particularly on a screen.

[0079] Furthermore, it should be mentioned that the procedure does not have to be carried out as analytically as described in the paragraphs above. In particular, a discriminant analysis can also be performed, optionally using machine learning algorithms, to determine which subgroup is most representative of the respective user.

[0080] Based on the above considerations, the most representative subgroup can be determined from the statistical properties. In the next step, a performance result for the first user within the selected subgroup is determined. This performance result could be, for example, the statement "X% better or worse than the average in this subgroup" or "Y seconds faster or slower than the average in this subgroup." This provides the user with information about their performance compared to other users within the same performance range. For example, the severity of symptoms in a patient who has recently suffered a stroke can be objectively determined. Furthermore, this performance result can be used to tailor physiotherapy and rehabilitation to the individual user.

[0081] So far, only the procedure for determining a performance result for the first performance measurement LMB 1-1 has been explained. However, as shown in Figure 1, a second performance measurement LMB 1-2 may also exist for the first user Bl. The aforementioned procedure can be repeated so that, regardless of the previous steps of determining the statistical property from the first performance measurements of the subgroups, statistical properties can be determined from the second performance measurements of the subgroups. This can be carried out using the same considerations as above; however, the procedure may result in a different subgroup being most representative for user Blam with respect to the second performance measurement LMB 1-2 than with respect to the first performance measurement LMB 1-1. This may be due to, for example, the fact that...A first exercise is difficult for a stroke patient to perform, but a second exercise can be performed to the same extent.

[0082] Alternatively, it can be provided that at least one statistical property for the subgroup is determined with respect to at least two performance measurements. In this case, a "simple" two-dimensional representation of distribution curves as in Figures 5 to 7 will not be possible; instead, the statistical property will be determined based on distributions in a multidimensional space. In other words, the statistical property is created as a function of the first and second performance measurements from the subgroups. Subsequently, the most representative subgroup for user B1 can be determined, analogous to the above, with the determination of the most representative subgroup taking place in a multidimensional space. In this case, in particular, discriminant analysis using machine learning algorithms is a suitable approach.

[0083] Figures 2 to 4 show in particular that the subgroups can also overlap, which means that identical DSi data sets can occur in at least two subgroups.

[0084] As explained above, the most representative subgroup for the user can be found using the method described above, allowing the user's performance to be compared with other stroke patients, for example. However, it is often desirable to also show the user their "target group," which, while not currently representative of the user (Bl), serves as a target template for the user to achieve. Figures 5 and 6 illustrate this. According to the method described above, the first subgroup was identified as the most representative, in which the user achieves a performance score of "better than 90% of stroke patients." The goal, however, is for the user to achieve above-average results in their weight class (60-80 kg) within the second subgroup, in which the first user (Bl) currently only achieves a performance score of "better than 40% of users in the 60-80 kg weight class."This serves as motivation for recovery, particularly in physiotherapy and rehabilitation. The two performance results can be displayed as a comparative representation on a screen. The aforementioned procedure can also be carried out for a second user, who initially provides their own personal data. The first user and the second user perform the same exercise simultaneously (or with a time delay) and thereby obtain one or more of their own performance measurements. The type of performance measurement for the first user corresponds to the type of performance measurement for the second user; that is, the same characteristic is measured, e.g., the time taken for a specific exercise or similar.The procedure described above is carried out independently for each user, so that a separate subgroup is found for each user, which is most representative of that user. This allows the performance results of the first and second users to be displayed, for example, as a comparative representation on a screen, showing how well each user performed compared to their most representative group of users.

[0085] After or even before the completion of the procedure, the data record DSB1 of the first user Bl shown in Figure 1 can be added to database 1 to extend it accordingly.

[0086] The examples above were explained specifically for the health sector, meaning that at least one piece of personal data relates to information about a user's illness or symptom. Database 1 comprises at least two, preferably over one hundred, records of individuals who have or have had the same illness or symptom as the first user, Bl. At least one further piece of personal data can be provided by the first user, Bl, and in the records DSi, indicating a period from the first occurrence of the illness or symptom, or from the period from the resolution of the illness or symptom. In these procedures, the severity of a symptom can be determined, with the severity of the symptom being the performance result of the first user's performance measurement within the subgroup considered.This result can be used to tailor physiotherapy or rehabilitation.

[0087] The described method could also be used in sports, particularly competitive sports, to determine an athlete's performance level. In this case, the database would contain at least two, preferably over one hundred, records of athletes participating in the same sport as the first user, Bl. Based on the performance result, which indicates the first user's performance level, a training plan can be developed.

Claims

Claims:

1. Computer-implemented procedure for performing a performance analysis of a first user (Bl), comprising the steps: - Providing a database (1) containing a plurality of data records (DSi), wherein each data record (DSi) includes a plurality of personal data (pD) and at least one initial performance metric (LM), - Receipt of personal data (pDB 1 - 1 , pDB 1 -2) of the first user (Bl) and a measured performance value (LMB1-1) of the first user (Bl) or an indication of a performance value to be measured (LMB1-1) of the first user (Bl), - Determining at least two distinct subgroups (UG1, UG2, UG3) of the database (1), wherein each subgroup (UG1, UG2, UG3) comprises only those records (DSi) of the database (1) where a selected date or a selected combination of at least two personal data (pD) matches or is related to a corresponding date or combination of personal data (pDBl-1, pDBl-2) of the first user (Bl), for the aforementioned subgroups (UG1, UG2, UG3), determine at least one statistical property, in particular a variance or standard deviation, from the performance measurements (PM) of the subgroups (UG1, UG2, UG3), - Selecting the subgroup (UG1, UG2, UG3) that is most representative of the first user (Bl) according to at least one determined statistical property, preferably by performing a discriminant analysis, - Performing the performance analysis of the first user (Bl) by determining a performance result of the performance measurement (LMB1-1) of the first user (Bl) within the subgroup used (UG1, UG2, UG3).

2. The method according to claim 1, wherein in the step of determining at least two different subgroups (UG1, UG2, UG3) all possible subgroups (UG1, UG2, UG3) are determined.

3. A method according to claim 1 or 2, wherein a statistical property is determined for a previously defined subgroup (UG1, UG2, UG3) only if the subgroup (UG1, UG2, UG3) is statistically relevant, in particular if the subgroup (UG1, UG2, UG3) comprises a predetermined number of data records (DSi).

4. A method according to any one of the preceding claims, wherein the method comprises the following steps: - Receiving a measured second power measurement (LMB1-2) from the first user (Bl) or receiving an indication of a second power measurement (LMB1-2) to be measured from the first user (Bl), Repeat the steps of identifying, gathering, and performing for the second power measurement (LMB1-2).

5. Method according to claim 4, wherein different subgroups (UG1, UG2, UG3) are used for the two power measurement values ​​(LMB1-1, LMB1-2), wherein the subgroups (UG1, UG2, UG3) particularly preferably overlap.

6. Method according to one of the preceding claims, wherein a further subgroup (UG1, UG2, UG3) is used and a further performance analysis of the first user (Bl) is carried out by determining a further performance result of the performance measurement value (LMB1-1) within the further subgroup used (UG1, UG2, UG3), wherein both performance results are preferably displayed as a comparative representation on a screen.

7. Method according to one of the preceding claims, wherein the steps receiving, determining, ascertaining, eliciting and performing are repeated for a second user, wherein the type of power measurement value (LMB1-1) of the first user (Bl) corresponds to the type of power measurement value (LMB1-1) of the second user, and wherein the elicited subgroups (UG1, UG2, UG3) are preferably different for the first user and the second user.

8. Method according to claim 7, wherein the performance measurements of the first user (Bl) and the second user are determined substantially simultaneously and the performance results of the first user (Bl) and the second user are preferably displayed as a comparative representation on a screen.

9. The method of claim 8, wherein, based on the performance result of the first user and the performance result of the second user, an overall ranking is created and displayed on the screen, in which the users are ordered according to their determined relative performance results.

10. The method of any one of the preceding claims, wherein the personal data (pDBl-1, pDBl-2) and the measured performance metrics (LMB1-1) of the first user (Bl) are added to the database (1) after the performance analysis has been carried out.

11. Method according to one of the preceding claims, wherein for the aforementioned subgroups (UG1, UG2, UG3), for which at least one statistical property has been determined, a representation factor is determined on the basis of the determined statistical properties, wherein the subgroup with the largest representation factor is determined as the most representative subgroup, wherein preferably at least one representation factor or at least two representation factors of at least two subgroups are output and particularly preferably displayed on a screen.

12. Method according to one of the preceding claims, wherein at least one of the determined statistical properties is a variance or standard deviation of the performance measurements of the subgroups and a subgroup is more representative the lower its variance or standard deviation.

13. Method according to one of the preceding claims, wherein at least one of the determined statistical properties depends on the performance measurement of the first user (Bl), where the aforementioned statistical property is preferably the distance of the performance measurement of the first user to the mean of the performance measurements of the subgroup, and a subgroup is more representative the smaller the distance of the performance measurement of the first user (Bl) to the mean of the performance measurements of the subgroup.

14. Method according to any of the preceding claims for determining the severity of a symptom, wherein the severity of the symptom is the performance result of the performance measurement (LMB1-1) of the first user (Bl) within the subgroup used (UG1, UG2, UG3).

15. A method according to any of the preceding claims for determining the performance level of an athlete, wherein the performance level is the performance result of the performance measurement (LMB1-1) of the first user (Bl) within the selected subgroup (UG1, UG2, UG3).

16. A method according to any of the preceding claims, wherein the determined performance result of the first user (Bl) is used to automatically create or adapt a training or therapy plan for the first user (Bl).

17. Method according to one of the preceding claims, wherein the performance measurement value is used for the automatic adjustment of at least one parameter of a computer game, in particular for adjusting a level of difficulty.

18. Method according to one of the preceding claims, wherein the performance result is provided as an input to a matchmaking algorithm, wherein the matchmaking algorithm makes an assignment between at least two players based on the performance result.

19. Method according to any of the preceding claims, wherein the at least one performance measurement is a measured value measured by a sensor, in particular a biomechanical or physiological measured value such as a force, a movement speed or a heart rate; or an exercise-specific or game-specific measured value such as a time required for an exercise or a game.

20. Method according to one of the preceding claims, wherein the power measurement values ​​are calculated from at least two power measurement data, which preferably are incorporated into the power measurement value with a weighting.

21. Device for data processing which is configured to carry out the method according to any one of claims 1 to 20.

Images