Predictive analytics method and system for positively adjusting fitness and/or well-being conditioning

Description

CROSS REFERENCE TO RELATED APPLICATIONS

This application claims the benefit of U.S. Provisional Application no. 62/018,783, filed Jun. 30, 2014, and entitled Artificial Intelligence Based System and Method for Athletic Performance Prediction and/or Beneficial Training Program Adjustment by the same inventors as above.

This application is related to U.S. patent application Ser. No. 13/912,176, entitled System and Method for Assessing Functional State of Body Systems Including Electromyography by Masakov, and filed on Jun. 6, 2013, which is hereby incorporated by reference as though disclosed herein. This application is also related to U.S. paatent pplication Ser. No. 13/912,178 entitled System and Method for Functional State and/or Performance Assessment and Training Program Adjustment by Nasedkin, and filed on Jun. 6, 2013, which is hereby incorporated by reference as though disclosed herein.

FIELD OF THE INVENTION

The present invention relates to a method and system for improving fitness, well-being and athletic performance—team or individual. More specifically, the present invention relates to an artificial intelligence (AI) based method and system for assessing multiple causal factors to a target outcome, determining their relationship and significance, and adjusting a conditioning or wellness program to achieve a more beneficial outcome. The present invention utilizes assessed bio-signals that are indicative of the current functional state of a user and predictive analytics.

BACKGROUND OF THE INVENTION

U.S. Pat. No. 6,572,558 was issued to Masakov, et al., for an Apparatus and Method for Non-Invasive Measurement of Current Functional State and Adaptive Response in Humans. This patent introduces the use cardiac, brainwave and related physiological signals to assess the current functional state of a user.

U.S. patent application Ser. No. 13/912,176 (noted above) discloses the inclusion of electromyography assessment in combination with other assessments to achieve a non-invasive, non-depleting comprehensive functional state assessment of a user.

U.S. patent application Ser. No. 13/912,178 (also noted above) discloses adaptive training including conducting a current functional state assessment prior to a conditioning procedure and adjusting the conditioning procedure in view of the current functional state of the user to enhance conditioning.

In addition to athletic and well-being conditioning, prior art related to the present invention may include predictive analytics and artificial intelligence models. Predictive analytics may include weather forecasting, financial market predictions, sports odds-making and other types of forecasting or predicting. Various methods are known and described in their respective literature.

Various artificial intelligence (AI) models and tools are known and include versions of search and mathematical optimization, logic, and methods based on probability and economics. Some problems in AI (in reasoning, planning, learning, perception and robotics) require the agent to operate with incomplete or uncertain information.

Bayesian networks are a general tool that can be used for a number of problems: reasoning (using the Bayesian inference algorithm), learning (using the expectation-maximization algorithm), planning (using decision networks) and perception (using dynamic Bayesian networks). Probabilistic algorithms can also be used for filtering, prediction, smoothing and finding explanations for streams of data, helping perception systems to analyze processes that occur over time (e.g., hidden Markov models or Kalman filters).

U.S. Pat. No. 8,620,852, issued to Kipersztok, for example, teaches a system and method to facilitate predictive accuracy for strategic decision support using Bayesian networks.

A problem in current fitness and well-being conditioning is that a large number of contributors or variables may influence a target outcome, both directly and indirectly. Furthermore, it is unknown, or difficult to ascertain, the significance of variables and their relationship to one another.

In addition, the prior art is disadvantageous in not presenting the functional state assessment parameters that are most significant to fitness and well-being conditioning and, in turn, the external contributors that are most significant in influencing the functional state assessment parameters. Without presentation of these variables a coach or user is limited in control and optimization of conditioning.

The present invention overcomes these shortcomings and meets an underlying, long-felt need—a need evidenced in part by the amount of money spent on coaching and conditioning for athletic performance and well-being.

SUMMARY OF THE INVENTION

Accordingly, it is an object of the present invention to provide a method and system that uses artificial intelligence or “predictive analytics” to positively adjusting fitness and/or well-being conditioning.

It is also an object of the present invention to provide a user or coach with mechanisms to readily assess and the many variable impacting a conditioning goal, how those variable interrelate and/or how modification of a given variable might impact a target outcome or another variable.

These and related objects of the present invention are achieved by use of a predictive analytics method and system for positively adjusting fitness and/or well-being conditioning as described herein.

The attainment of the foregoing and related advantages and features of the invention should be more readily apparent to those skilled in the art, after review of the following more detailed description of the invention taken together with the drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

FIGS. 1-2 are a solution diagram representing a target outcome (TO), identification of significant contributors to the outcome, and the evolving solution.

FIG. 3 illustrates one embodiment of a solution dashboard for the contributors to the solution of FIGS. 1-2.

FIG. 4 is a flow diagram of general processing in accordance with the present invention.

FIG. 5 is a diagram that shows the unconnected variables that have been selected as potential contributors to the target outcome.

FIG. 6 illustrates the unconnected data of FIG. 5 is processed with a predictive ANB model for Win/Loss(P) to create an initial predictive model for Win/Loss.

FIG. 7 is an ANB model diagram for investigating ICs contributing to Hours Traveled.

FIG. 8 is an ANB model for external contributors that may influence the selected internal contributor (LF nu).

FIG. 9 is a diagram of an ANB model to determine the internal contributor influencing gym sets and reps.

FIG. 10 is a solution diagram with the internal and external contributors for Hours Traveled and Gym SetsxReps.

FIG. 11 is a diagram of normalized mutual information between variables this is instructive in the selection of muscle soreness as an external contributor.

DETAILED DESCRIPTION

Definitions:

For purposes of teaching and claiming the present invention in accordance with 35 U.S.C. section 112, the following terms are defined generally as follows and in a manner consistent with their use in the patents and patent applications referred to herein.

“Current Functional State” refers to the physiological state of a user (typically readiness for physical activity, though may be another depending on Target Outcome) as indicated by the functional state assessment discussed herein, which may include cardio, brainwave, muscular and other assessments or combinations thereof, particularly of the typed described in U.S. patent application Ser. Nos. 13/912,176 and 13/912,178, and listed as internal contributors.

“Current Functional State Assessment” (CFSA) refers to a physiology-based assessment that determines the Current Function State of a user. A CFSA may involve one or more of the assessments referred to in the Current Functional State definition above, and typically includes one, some or all of the internal contributor assessments. Each assessment may be indicated with an index value. The CFSA may give an indication of the “readiness” of the user for physical work.

“Target Outcome” (TO) refers to a selected physiologically related goal or outcome for which the contributors to that outcome are being investigated and determined by the present invention. The target outcome may be winning an athletic match, scoring a number of points, completing a particular type of race in a certain time, losing weight, reducing cardio-vascular build up, or other physiological-based goal or desired outcome.

“Primary Contributor” (PC) refers to a parameter, variable or contributor that has a causal, determinative or influencing effect on the Target Outcome.

“Internal Contributor” (IC) refers to a physiological CFSA parameter, variable or contributor, such as Adaption Response, HR at AnT, Aerobic index, LF nu, or another one of the many listed below in definitions or listed in the chart of FIG. 5. The internal contributors have a causal, influential or determinative relationship to the value of a Primary Contributor.

“External Contributor” (EC) refers to a parameter, variable or contributor that has a causal, influential or determinative relationship to the index value of an Internal Contributor. External contributors may be training or non-training factors.

Parameter/Contributor Definitions:

The following definitions include name, definition, symbol and unit of measure. These are variables that may be primary contributors, internal contributors or external contributors to a chosen target outcome. Note that the word parameter and contributor are used rather interchangeable in the discussion below.

• 1. Stress index—The level of tension in the cardiac system in response to physical and mental loads. SI. Relative Units /7 (i.e., range of 1-7).
• 2. Fatigue—The temporary state of the cardiac system that occurs during intense or prolonged work that leads to a decrease in the effectiveness of the work. FT. Relative Units /7.
• 3. Adaptation Reserves—A measure of how long and effectively the cardiac system can express the ability to adapt to external stimuli. AR. Relative Units /7.
• 4. Central Nervous System—Level of activation and intensity of functioning of the Central Nervous System at a specific moment in time. A comprehensive indicator of the current state of the Central Nervous System presented on a scale of 1 to 7; 1 being a very poor state of Readiness and 7 being an excellent state of Readiness. The current state of the Central Nervous System determines its ability to effectively regulate the functions of the body in order to achieve useful adaptive results from training. CNS. Relative Units /7.
• 5. Aerobic Status Index—Reflects the current state of the aerobic metabolic pathway and the ability to perform aerobic work in training. ASI. Relative Units.
• 6. Anaerobic Status Index—The current ability to perform muscle work using the glycolytic energy system while withstanding a high level of lactate in the blood. AnSI. Relative Units.
• 7. Heart Rate at Anaerobic Threshold—The key component that defines the efficiency of energy production during prolonged exercise, as well as being an indicator of the athlete's overall level of endurance. HRAT. Beats Per Minutes.
• 8. Metabolic Reaction Index—Reflects the overall effectiveness and coordination of the metabolic system to support planned training loads, measured over longer periods of time. MRI. Relative Units.
• 9. Metabolic Reaction Index Grade—The Metabolic Reaction Index Grade characterizes the overall metabolic capacity of both the aerobic and anaerobic mechanisms of energy supply to support any kind of physical exertion. MRIG. Relative Units /7.
• 10. Metabolic Grade—Reflects overall metabolic system's readiness. MG. Relative Units.
• 11. Activity of Vagus Regulatory Mechanisms—Indicates the current activation level of the parasympathetic nervous system's regulation of the cardiac system, which serves to maintain homeostasis and restore the functionality of the body after load. AVRM. Sec.
• 12. Activity of Sympathetic Regulatory Mechanisms Indicates the current activation level of the sympathetic nervous system's regulation of the cardiac system. ASRM. %.
• 13. Tension Index—The level of tension in the cardiac system in response to physical and mental loads. Reflects the level of centralization of heart rhythm regulation. Centralization involves increased involvement of central levels of regulation and a decreased level of autonomic regulation of heart rhythm. TI. Relative Units.
• 14. Standard Deviation of the Aspirate Waves—Reflects the level of automatization of heart rhythm regulation.

Automatization involves a predominance of autonomic regulation and a decreased responsibility of central levels of regulation. SDAW. Relative Units.

• 15. Share of the Aperiodic Influences—Reflects the level of random and aperiodic activity that influences heart rhythm. Slow waves reflect activation of the central circuit and a predominance of activity in the cardio-stimulatory center (which is part of the medulla). Ex: over-trained athletes express slow waves in a significant manner. SAI. Sec.
• 16. Standard Deviation Normal to Normal—Standard deviation of the full array of cardio intervals. Reflects the total effect of autonomic regulation. SDNN. ms.
• 17. Standard Deviation of Successive Differences—Standard deviation of differences between adjacent normal to normal intervals. SDSD. ms.
• 18. Root of the Mean Square of the Difference of Successive intervals—The square root of the sum of differences of a sequential series of cardiointervals. Reflects the parasympathetic activity. RMSSD. ms.
• 19. Total Power—Variance of all normal-to-normal intervals, ≤0.4 Hz. Reflects the level of activity of regulatory system. TP. ms2.
• 20. Low Frequency/High Frequency Ratio—Ratio between low and high frequency components. Reflects the Sympathetic-Parasympathetic Balance. Ratio LF ms2/HF ms2. LF/HF.
• 21. High Frequency—Power in high frequency range 0.15-0.4 Hz. Reflects the parasympathetic activity. HF. ms2.
• 22. High Frequency (Normalized Units)—HF power in normalised units. HF/(Total Power−VLF)×100, n.u. HF n.u. Normalized Units.
• 23. Low Frequency—Power in low frequency range 0.04-0.15 Hz. Reflects the sympathetic activity. LF. ms2.
• 24. Low Frequency (Normalized Units)—LF power in normalised units LF/(Total Power−VLF)×100, n.u. LF n.u. Normalized Units.
• 25. Very Low Frequency. Power in very low frequency range ≤0.04 Hz. Reflects the humoral activity. VLF. ms2.
• 26. Omega resting potential—The present activation level of the frontal brain's systems that comprise the integrative center. DC. mV.
• 27. Direct Current Potential Stability—A qualitative analysis of the resting potential's form or shape. Categorized as either Stable or Unstable. DCS. Stable or Unstable
• 28. Direct Current Potential Variation to Stabilisation—The difference between initial resting potential values (in the beginning of the measurement) and final values (at the end of the measurement). DCVS. mV
• 29. Direct Current Potential Time to Stabilisation—The duration required to reach stabilisation during the measurement. DCTS. Sec.
• 30. Days Post Game—This signifies the amount of days post game that the Omegawave assessment was taken. DPG. Days.
• 31. Hours Post Last Exercise—This signifies the amount of hours post exercise that the Omegawave was taken. If a player trained at 6 pm, yet undertook an Omegawave assessment at 8 am the next day, this would mean the assessment took place 14 hours post exercise. HPLE. Hours.
• 32. Days Until Game—This signifies the amount of days prior to the next game from when the player took the CFS assessment. If a player undertook an assessment on Wednesday and the next game was Saturday, the days until next game would be three. DUG. Days.
• 33. Age—Athlete's (or user's) age. A. Years.
• 34. Weight—Athlete's body mass. W. kg.
• 35. Readiness to Train/50—This related to a players self-report wellbeing/readiness questionnaire for that given day were readiness to train is assessed from five indices; energy levels, muscle soreness, motivation, sleep quality and appetite. These five indices act as a scale upon which players rate their individual wellbeing for the day—this score is out of 50, with 10 being very high well being/readiness and 1 being low well being/readiness for the five indices outlined below. A score out of ten is taken from the five indices below to give a readiness to train score out of 50. RT. Relative Units /50
• 36. Energy Levels /10—This related to a players self-report wellbeing/readiness questionnaire for that given day were energy levels is one of five indices upon which players rate their individual wellbeing—this score is out of 10, with 10 being very motivated and 1 being not motivated. EL. Relative Units /10.
• 37. Muscle Soreness /10—This related to a players self-report wellbeing/readiness questionnaire for that given day were muscle soreness is one of five indices upon which players rate their individual wellbeing—this score is out of 10, with 10 being experiencing high muscle soreness being a 1 and were a player experiences no muscle soreness being 10. MS. Relative Units /10.
• 38. Motivation /10—This related to a players self-report wellbeing questionnaire for that given day were motivation is one of five indices upon which players rate their individual wellbeing—this score is out if 10, with being very motivated and 1 being not motivated. M. Relative Units /10.
• 39. Sleep Quality /10—This related to a players self-report wellbeing/readiness questionnaire for that given day were sleep quality is one of five indices upon which players rate their individual wellbeing—this score is out of 10, with 10 being experiencing poor sleep quality being a 1 and were a player experiences good sleep quality being 10. SQ/10. Relative Units /10.
• 40. Appetite /10—This related to a players self-report wellbeing/readiness questionnaire for that given day were appetite is one of five indices upon which players rate their individual wellbeing—this score is out of 10, with 10 being good appetite (hungry) and 1 being poor appetite (not hungry). App. Relative Units /10.
• 41. Sleep Quantity—This is the amount of hours sleep a player had the night prior to an Omegawave assessment. This score is not included within the self-report wellbeing/readiness questionnaire score but is obviously related to the sleep quality. SQ. Relative Units.
• 42. Metres Covered—This value is the amount of metres the player covered in the day immediately prior to CFS assessment. This value is taken from StatSports Viper GPS units worn by the players on the training field or during games. Metres. Metres
• 43. Dynamic Stress Load—This value is dynamic stress load in the day immediately prior to CFS assessment. This value is taken from StatSports Viper GPS units worn by the players on the training field or during games. DSL was noted as the “total of the weighted impacts, which is based upon accelerometer values of magnitude above 2G” (StatSports Viper Metrics, 2012, p. 3). DSL. Relative Units.
• 44. High End Impacts—This value is high end impacts in the day immediately prior to CFS assessment. This value is taken from StatSports Viper GPS units worn by the players on the training field or during games. The StatSports device, which includes a 100 Hz 3-D accelerometer and measures GPS impacts when values were above 2G in a 0.1 second period. The weighting of GPS impacts are totalled and scaled to provide workable values, with the impact of 4G being twice as hard on the body than a 2G impact. It is important to note that GPS impacts are a combination of collision and impacts created from movement (stepping, jumping etc.). HEI. Relative Units.
• 45. Game Minutes Played—This value is the amount of minutes played by a player in game situations against opposition. A rugby games lasts 80 minutes and players can be subbed on and off at any time. For example if a players played 80 minutes on Saturday and an CFS assessment took place two days later on a Monday this minutes played was added. GMP. Minutes.
• 46. Win/Loss—This signifies if the team the player was representing won or lost in the most recent game. Win or loss is noted in relation to a player's most recent CFS assessment taking place. For example if a players team lost a game on Saturday and an CFS assessment took place two days later on a Monday this won/lost was added. W/L.
• 47. Hours Travelled—The number of hours travelled by the player in the 48 hours prior to CFS assessment. HT. Hours.
• 48. Total Tackles Completed—In a game situation were a player successfully completes a tackle. The greater the amount of tackles completed by a player the better. TT. Relative Units.
• 49. Dominate Tackles—In a game situation were a player successfully completes a dominant tackle, whereby they force the opposition backwards. The greater the amount of dominant ball carries by a player the better. DT. Relative Units.
• 50. Total Tackles Missed—In a game situation were a player does not successfully complete a tackle. The fewer total missed tackles by a player the better. MT. Relative Units.
• 51. Tackle Completion (%)—In a game situation were a player successfully completes a tackle and the number of successful tackles is compared against the unsuccessful tackles to provide a percentage completion. The greater the value of tackle completion by a player the better. T %. Relative Units.
• 52. Ball Carries—In a game situation were a player carries the ball in an attacking situation into contact with the opposition. The greater the amount of ball carries by a player the better. BC. Relative Units.
• 53. Gainlinie Plus Success (%)—In a game situation were a player carries the ball in an attacking situation into contact with the opposition and forces the opposition backwards. This figure can be calculated as a percentage when used in conjunction with instances when a player carries the ball in an attacking situation into contact with the opposition and does not force the opposition backwards. The greater the amount of gainline success by a player the better. G+. Relative Units.
• 54. Effective Ball Presentation (%)—In a game situation were a player presents the ball correctly after a ball carry. The greater the amount of effective ball presentations by a player the better. BP. Relative Units.
• 55. Ruck Clearance Efficiency (%)—In a game situation were a player enters a “ruck” situation effectively. The greater the amount of effective ruck clearances by a player the better. R. Relative Units.
• 56. Total Turnovers Conceded—In a game situation were a player concedes possession of the ball. The fewer total turnovers conceded by a player the better. TTC. Relative Units.
• 57. Penalties Conceded—In a game situation were a player concedes a penalty. The fewer total penalties conceded by a player the better. TP. Relative Units.
• 58. Total Training Load—This signifies the training load for the training week to date, using the Borg system detailed below. If a training week started on Monday and CFS assessment was taken on Friday, this training load value would include any days were the player had trained or played in the days from Monday to prior to the CFS assessment using the Borg RPE scale for each day. Weekly TL. Relative Units.
• 59. Training Load Min×RPE—Using the Borg RPE scale the amount of time (mins) spent training by the individual in a session is multiplied by the given RPE for that session to provide a training load score. Session TL. Relative Units.
• 60. Subjective Training Load—This signifies the subjective training load given to the players training and playing schedule for the previous three days prior to assessment. This can be rated light, moderate or heavy depending upin the exertion perceived by the coach in the three days assessed. Subjective TL. Relative Units.
• 61. Sets×Reps Gym—A score derived from the gym session undertaken immediately prior to CFS assessment were the number of gym receptions are multiple by the number of sets—for example 10 sets of 10 equals 100. S×R. Relative Units.
• 62. Gym Focus—The predominant gym focus for the session undertaken prior to CFS assessment—Hypertrophy-Strength-Power. Gym. Relative Units.
• 63. SAQ—This signifies were players undertook speed, agility or quickness sessions in the day prior to CFS assessment. This value is noted in minutes for the amount of time designated to SAQ. SAQ involves many explosive movements and power elements of training that can be conducted on the training field or in the gym. SAQ. Relative Units.
• 64. Maximal Aerobic Speed—A form of aerobic training were time spent above 100% of maximal aerobic speed enables improvement of aerobic power—this training method is popular in team sport setting as the intensity involved in interval work produces the supra maximal training impulse. MAS. Relative Units.
• 65. Cryotherapy—This signifies were a player undertook whole body cryotherapy treatment in the day prior to CFS assessment taking place. This was undertaken in a cryotherapy chamber and involved exposure up to −120 degrees for three minutes. This whole body cryotherapy was only completed on selected training days at varying times of the day. WBC. Relative Units.
• 66. Ball Presentation—In game situation, were a player presents the ball after a carry. BP. %

The present invention includes machine executable software that executes on a computing device, which may be a computer (mainframe, desktop or laptop), a tablet, a mobile phone or other mobile computing device, a watch or other wearable computing device. The make up of these computing devices and the execution of software thereon are well known in the art.

Referring to FIGS. 1-2, a solution diagram 10,110 representing a target outcome (TO), identification of significant contributors 11-13,21-23,31-33,111-113,121-123,131-133 to the outcome 5,105, and the evolving solution (evolving through artificial intelligence) in accordance with the present invention is shown. In FIG. 1, the contributors are not yet determined and are represented by blank boxes. In FIG. 2, the contributors have been identified. The present invention includes the steps of identifying the contributors (and their significance/weighting) and continually evolving or optimizing the solution. The present invention also includes providing tools, awareness and access to the user or his/her coach to control and select optimization approaches (i.e., the solution).

The TO may be selected as winning in an athletic competition, improving personal fitness, losing weight, addressing a health concern, or any other physiological related objective that can be impacted by conditioning (i.e., by exercise, diet, sleep, physical therapy, massage, and/or other fitness or well-being considerations).

In a first example (shown in FIG. 2 and discussed in more detail below), the target outcome 105 is selected as winning for a given professional rugby team. Using the method and system of the present invention, it is possible to determine factors that are primary contributors (PCs) 111-113 to win/loss. These PCs are (limited to the present example): Hours Traveled, Work Capacity and Ball Presentation. It should be noted that these factors may change or be different over time and for different teams/individuals, and for different target outcomes, etc.

As deduced by the inventors herein, these primary contributors are in turn influenced by physiological contributors, termed “internal contributors” 121-123 that a coach, athlete, or other user may be able to investigate, account for, manipulate, and/or improve through conditioning and beneficial adaptation. These internal contributors include current state assessment parameters such as those listed in the Contributor Definitions above, and related parameters, including other parameters devised in the future. The generation and use of internal contributors is discussed in U.S. patent applications Ser. Nos. 13/912,176 and 13/912,178, referenced above, among other sources.

The internal contributors are in turn influenced by external contributors 131-133. External contributors may include type, intensity and volume of training as well as many other contributors including, but not limited to several of the contributors defined above and/or discussed herein. Note that the type and breadth of external contributors may vary widely depending on the target outcome and user(s) involved.

Once the contributors are identified, a user or coach may begin to modify or address the multiple contributors. This integration of multiple contributors is termed the “conclusion” 40,140 and from it the optimized “solution” 50,150 is implemented. The monitoring of the solutions efficacy and the feeding back of those data points to the model that create the contributors and their weighting is termed the “intervention” 60,160.

Referring to FIG. 3, one embodiment of a solution dashboard 170 in accordance with the present invention is shown. The present invention permits a user or coach to select the contributors the user/coach would most like to readily see. These are assembled on to the dashboard. Nine contributors 171-179 are shown on dashboard 170, and these include ICs and ECs, with the ECs including intensity (rate, frequency, etc.) and volume (distance, amount lifted, etc.) of exercise. While nine contributors are shown in FIG. 3, more, less, or other contributors may be displayed.

The dashboard permits a user/coach to quickly see the level of various contributors and their likely cumulative influence on success (achieving of the TO), indicated by gauge 180. Also, in a solution-exploratory mode, the dashboard permits the user/coach to virtually experiment with various conditioning steps. For example, the coach may move the intensity of exercise bar 175 and see how that influences the other parameters. Similarly, the volume bar 179 can be slide to see its effect. While the ICs are not directly selectable by a user/coach, they too can be adjusted exploratorily. If a beneficial level is found, the user/coach can attempt to adjust other aspects of training and/or lifestyle to achieve the desired beneficial level.

Thus, among other benefits, dashboard 170 permits a user/coach to see current levels and the interrelation of those levels with one another. In addition, the user/coach can forecast or craft training plans by moving one or more of the values and seeing how that change affects/influences the other parameters. This permits a user/coach to optimize workouts and conditioning or at least knowingly take steps in the direction.

Referring to FIG. 4, a flow diagram of general processing in accordance with the present invention in shown. The process begins with data collection, step 184. This may be of parameters that are anthropometric, physiologic, training loads, training types, player profiles and experience, and performance. Examples include those listed above and/or shown in the figures or mentioned elsewhere.

The present invention involves, in part, a mixing of structure machine-based learning and human based knowledge and selection. For example, there is human knowledge that is key to selecting the appropriate initial group of parameters, yet machine knowledge that may determine their level of influence (and recalculate using fewer and more prominent parameters). In general, there is an iterative and integrated process between human knowledge and selection and machine based knowledge and learning.

Once the desired data is collected, it is preprocessed, step 185. This may include “cleaning” the data or transforming (e.g., normalizing) it to a range or value of an appropriate magnitude and/or quality. This may be followed by an appropriate ordering and, in some instances, aggregation. Aggregation may refer to combining parameters into a “new” parameter to reduce parameter number and subsequent processing. Representative data parameters are listed in the definitions section above appropriate units.

This data is then preferably reviewed for missing values, correct format, and/or other irregularities. Discretization may be performed to prepare data values for subsequent processing. Next, new attributes may be constructed.

The “pre-processed” data is then “prepared” for analysis and processing, step 186. This may include completing or accumulating data sets, building data tables, adjusting algorithms, and other steps in data preparation for neural and Bayesian network analysis.

The next step 187 in process flow is “analysis and processing” which may include machine learning (preferably in a Bayesian manner) leading to determination and description of the parameters and outcome, relation of parameters to outcome and other parameters, prediction of outcome, and optimization of parameter selection and processing.

The analysis and processing step also include executing “reporting tools” that generating reports, and building dashboard 170, etc. These reports may vary yet may include a ranking of the influence of various contributors on target outcome and the magnitude of that influence, information on the influential parameter type and its value, and other information desired by an athlete, coach or trainer for adjusting conditioning and other decision points to improve or achieve the target outcome.

Bayesian Networks & Data Collection

Artificial intelligence in a preferred method of the present invention is carried out using an augmented naive bayes (ABN) method. The term “Bayesian networks” was coined by Judea Pearl in 1985 to emphasize three aspects: the often subjective nature of input information; the reliance on Bayes' conditioning as the basis for updating information; and the distinction between causal and evidential modes of reasoning.

A Bayesian network, Bayes network, belief network, Bayes(ian) model or probabilistic directed acyclic graphical model is a probabilistic graphical model (a type of statistical model) that represents a set of random variables and their conditional dependencies via a directed acyclic graph (DAG). For example, a Bayesian network could represent the probabilistic relationships between diseases and symptoms. Given symptoms, the network can be used to compute the probabilities of the presence of various diseases.

Formally, Bayesian networks are DAGs whose nodes represent random variables in the Bayesian sense: they may be observable quantities, latent variables, unknown parameters or hypotheses. Edges represent conditional dependencies; nodes that are not connected represent variables that are conditionally independent of each other. Each node is associated with a probability function that takes, as input, a particular set of values for the node's parent variables, and gives (as output) the probability (or probability distribution, if applicable) of the variable represented by the node.

The Naive Bayes model is a special form of Bayesian network. This model is mainly used for classification problems. The important feature of Naive Bayes model is that it has very strong independence assumptions.

As stated above, the Augmented Naive Bayes (ANB) is preferably used in the present invention, including in predicting achievement of the target outcome from available parameters. ANB is preferred for at least three reasons: (1) it fits well for classification tasks with a low number of samples, (2) high number of parameters and (3) potential multicollinearity between the parameters, as is the case in the used dataset.

ABN models are generated and they are preferably validated with help of Area Under ROC Curve (AUC)—metrics and Confusion Matrix and by using cross-validation with K=3 (K-Fold). To determine the predictive power of explanatory variables, Normalized Mutual Information (NMI) is preferably used to describe the relative influence of explanatory variables on success (achieving target outcome). In addition, hierarchical clustering is preferably used to identify the components of athlete preparation and their relation to desired target.

Initial data collection and processing includes the step of filtering out extra parameters not relevant in the prediction. They may include 1) parameters with deterministic relationship with other variables and 2) parameters with 35% or more missing values. A next is to find relevant categories and intervals for the classification process. K-means, based and manual discretization, were used in this.

In gathering/selecting data, parameters are gathered that may impact the selected “target outcome.” These parameters may be categorical or numerical or other, and may contain anthropometrical, physiological, performance, injury, training load or other variables. In addition, dataset may contain survey data, such as sleep quality and quantity, motivation and appetite as well as game statistics, and other data.

The physiological readiness of the Central Nervous System (CNS), Cardiovascular System (CVS), Energy Supply System (ESS) and/or other physiological parameters, i.e, the internal contributors, are preferably monitored and assessed frequently. Thus, a non-invasive, non-depleting assessment protocol, as taught by the referenced patent applications is preferred.

The present invention will now be further taught through example.

EXAMPLE 1

This study involves 14 players of a professional rugby team during the 2013-14 season. Target Outcome: winning league matches. The data contained anthropometrical, physiological, performance, training load and self-reported wellbeing variables. Physiological readiness of the Central Nervous System, Cardiac System (CS) and Autonomic Nervous System (ANS) were frequently monitored by Omegawave Team+ (Finland). The analytical logic with Primary, Internal and External contributors' identification was used. Supervised Learning—Augmented Naïve Bayes (ANB)—was used to predict targets. Irrelevant parameters were excluded based on their deterministic relationship with other variables. Predictive accuracy of the models was confirmed with the area under the Receiver Operating Characteristic curve (ROC index) and with a Confusion Matrix of cross-validation (K-Fold) K=3. Normalized Mutual Information (NMI) was used to determine the relative influence of variables on the target.

Referring to FIG. 5, a diagram is presented that shows the unconnected variables that have been selected as potential contributors to the target outcome. These include functional state measurements (non-game data) such as adaptation reserves and HR at AnT; actual game data with addition (F) for future game compared to non-game data (measuring during game or game trip) such as ball carries(F) and SAQ(F); actual game data with addition (P) for past game compared to other non-game data such as ball carries(P) and SAQ(P); and non-game data dedicated to non-functional state measurements such as age and hours post last exercise.

Referring to FIG. 6, the unconnected data of FIG. 5 is processed with a predictive ANB model for Win/Loss(P) to create an initial predictive model for Win/Loss. From this model, significant direct predictors, in other words, the primary contributors, can be identified. For validation, the model is subject to a ROC and K-Fold assessment. If valid, the NMI and p-value may be used (among other factors) to assess significance. Using cross-validation method based on K-Fold, total precision was 88.98%. Average ROC (Area Under ROC Curve—AUC) with cross validation and K=3 is 93.5%, which indicates a very high level of validity. The ROC is very high because the parameters are working together.

Using a target analysis, correlation with target node, eleven direct and significant predictors (primary contributors) were identified. The top three were Hours Traveled, Gym Reps, and Ball Presentation with NMI % (how much known about the Target) of 11.47%, 7.78% and 6%, respectively. These three were also most significant based on p-value. A mutual information and binary mutual information analysis (part of correlation with target node) and conditional mean analysis also confirmed the significance and supremacy of these contributors. Note that the ANB model of FIG. 6 could be redone (run a second time, or more, time) with a more refined set of contributors, ie., the more relevant contributors to further validate the model.

Having established these three primary contributors, a next step is the determination of which internal contributors influence each primary contributor. Note that one, two, four or another number of primary contributors could be selected (assuming the selected number of contributors are available in the dataset).

Referring to FIG. 7, an ANB model diagram for investigating ICs contributing to Hours Traveled is shown. FIG. 7 illustrates the relationship of the functional state measurements to the target node (outcome) and to one another. This model reflects the total effect of each functional state variable on Hours Traveled. ROC for this model is 85%.

For this step, the target variable is preferably considered to be locally linear and the total effect is the estimation of the derivative of the target with respect to this variable. The total effect represents the impact of a small modification of the “mean” of a variable over the “mean” of the target. The “total effect” is the obtained ratio. In addition, a standardized total effect (STE) corresponds to the total effect multiplied by the ratio to the standard deviation of the current variable and the standard deviation of the target. LF nu demonstrated the highest STE, significantly (p<0.05) compared to other variables. This internal contributor (LF nu) reflects the activity and reserve of the sympathetic nervous system. A mutual information and binary mutual information analysis, and a conditional mean analysis also confirmed the determination of LF nu.

ANB model permits use of tools such as target mean analysis (standard effect and direct effect). This analysis allows graphical representation of the impact of changes in the selected nodes' means on the target node's mean. This permits the relationship between each node and the target variable to be viewed in the form of curves. Some of these curves may have regions of collinearity.

Multicollinearity is a statistical phenomenon in which two or more predictor variables in a multiple regression model are highly correlated, meaning that one can be linearly predicted from the others with a non-trivial degree of accuracy. In this situation the coefficient estimates of the multiple regression may change erratically in response to small changes in the model or the data. Multicollinearity does not reduce the predictive power or reliability of the model as a whole, at least within the sample data themselves; it only affects calculations regarding individual predictors. That is, a multiple regression model with correlated predictors can indicate how well the entire bundle of predictors predicts the outcome variable, but it may not give valid results about any individual predictor, or about which predictors are redundant with respect to others. In case of perfect multicollinearity the predictor matrix is singular and therefore cannot be inverted. This analysis allows approximating the intervention (60,160) rapidly with reasonable accuracy

Referring to FIG. 8, a ANB model is shown for external contributors that may influence the selected internal contributor (LF nu). 19 potential contributors were included for analysis, 3 discrete, 16 continuous. With the model ROC is low, 64%. In this and similar instances an additional parameter may be pasted to the model to improve accuracy. When adding data in improve accuracy it is preferred to add data from multiple classes or clusters, for example, performance, anthropomorphic (age, height, etc.), physiologic, etc., and have data of good quality. This new data may be added to the model and ROC re-assessed.

Of the external contributors, there are three—Appetite, Readiness to Train, and Subjective Training Loads—that are close in their influence on LF nu. One or more subsequent analyses may be used to facilitate contributor selection in these instances. Those analyses may include, but are not limited to binary mutual information, mutual information, conditional mean, TE (total effect), STE, and target means (standard and/or direct) analyses. In the present case, looking at a cumulative significance over several analysis tools in helpful. Of those, correlation with target node, particularly binary mutual information, may be helpful. Also, total effect analysis is may be particularly helpful.

Referring to FIG. 9, a diagram of an ANB model to determine the internal contributor influencing gym sets and reps is shown. The ROC for this model is 83%, 90%+ being excellent and 80-90% being good. Validation, correlation and analysis was carried out as described above for the preceding contributors. DC was selected as the most influentially relevant internal contributor. Correlation to target node analysis supported this selection.

A next step is determining the external contributor that is most relevant or influential to DC. An ANB model was implemented with DC as the target node. This returned a low ROC or validation score. The ROC score can be improved by adding additional data as discussed above.

The analyses discussed above were conducted with the external contributor data of the model. Motivation, Muscle Soreness and Appetite all appeared to have a notable influence. Motivation was selected as the “best fit” or most influence contributor in view of these analyses, including the mutual information and target means analyses.

Referring to FIG. 10, a solution diagram with the internal and external contributors for Hours Traveled and Gym SetsxReps is shown. This indicates that LF nu and Appetite and DC and Motivation are the respective internal and external contributors.

A similar process in undertaken for the contributors to Ball Presentation. Completion of these steps yields the solution diagram of FIG. 2, with MRI (metabolic reactive index) and muscle soreness being the selected internal 123 and external 133 contributors, respectively.

Referring to FIG. 11, a diagram is shown that was useful in selection muscle soreness for external contributor 133. This diagram illustrates normalized mutual information between variables. The percentage shown is the contribution of that variable to the target node. Muscle soreness is the highest percentage. Note that a Pearson correlation is also helpful in making this selection and in other related selections herein.

Results

The initial ANB model for game success included 59 parameters. Predictive accuracy was excellent; average ROC index was 93% computed by cross-validation. Eleven significant predictors were found (p<0.1). Three Primary direct predictors were: Hours Travelled (NMI=11.5%); Resistance Training Load—Sets×Reps (NMI=7.8%); Training Orientation (NMI=7.2%). Internal (physiological) contributors that could improve the Primary contributors (i.e.: Hours Travelled) were identified. An ANB model for Hours Travelled with 26 physiological parameters was created (ROC=84%). Two of the most important contributors involved the CS and ANS: Low Frequency (p<0.05) and High Frequency in Normalized Units (p<0.1). Certain External contributors (i.e. Nutrition and Training Load) were found to be the most significant for optimizing Internal contributors (i.e. CS and ANS), thereby improving a player's trainability. Using predictive biological modeling, an optimal roadmap to successful performance was developed.

While the invention has been described in connection with specific embodiments thereof, it will be understood that it is capable of further modification, and this application is intended to cover any variations, uses, or adaptations of the invention following, in general, the principles of the invention and including such departures from the present disclosure as come within known or customary practice in the art to which the invention pertains and as may be applied to the essential features hereinbefore set forth, and as fall within the scope of the invention and the limits of the appended claims.