Svexa tackles a wide range of client needs
With our combined approach of consulting, algorithm licensing and technology solutions, we're able to meet the needs of different types of clients and partners. These examples cover our work in professional sports, with elite athlete coaches, and with individual athletes in a variety of disciplines.
Advising National Swedish Women’s Handball Team
SOCCER: WERDER BREMEN
Injury Prediction, Player Passports, and Optimized Data Handling
Challenge: The well known and historically very successful Bundesliga club SV Werder Bremen realized that they had lost ground to their competitors. They teamed up with betahausX to create The WERDER LAB with the goal of finding innovations from all around the world, combining sports with technology, that could benefit the athletes in terms of talent prediction and injury prevention.
Solution: We pitched our comprehensive and integrated data analysis, individual algorithms and profiles for all players, and injury prediction. Additionally, our team’s strong scientific ability and knowledge in applied sports were aspects considered positive.
Result: We were successful first in being selected among the 10 % that reached the final and then also in winning. SVEXA has since March been working with the Sports Performance Department at Werder Bremen. The exact data and results are obviously secret, but we are very happy with the accuracy of our algorithms, with injury prediction up to over 75 % (for players that have been injured several times) with really low false-positive rate below 5 %
Training Optimization for Elite Swimmers
Challenge: Training planning is an art with an almost infinite number of variables contributing to the final result. There are many tests and monitoring techniques, and all decisions are made in tight collaboration between coach and athlete. A big problem is that you only get very few chances at the biggest international competitions, and consequently, you get very few chances of optimizing the training and tapering plan.
Solution: Our development of the athletes’ individual profiles (below) enables a situation where we can use the coach’s initial training plan, run through our algorithms, and simulate many different adjustments, training plans, and tapering models - picking the most appropriate to execute.
Result: In our first test over three months, the coach’s initial response was that the analysis and data seemed reasonable, for example in terms of higher and lower loads on certain weeks, and adjustment in length of tapering. The first batch (Nov 2019-Feb 2020) was extremely successful, including world-class swimmers reaching new personal bests by over 1.00 sec!
The work with the Swedish National team using these algorithms to optimize training and performance has been intensified towards the now 2021 Tokyo Olympics.
DOWNLOAD CASE STUDY (PDF)
PERSONALIZED AI-DRIVEN COACHING
Adaptive, elite-level, automated training programs for recreational athletes
Challenge: Most recreational athletes do not have access to any type of coaching. This results in the use of ad-hoc or generic training programs that are neither personalized or adaptive over time. This approach on average leads to significant underperformance and even injury.
Solution: To address this need, we combined several of our products to create the training planning system Ellida*, which combines training logs, subjective measures (fitness level, goals, and commitment) and coaching domain expertise to produce then iterate on personalized training plans. In partnership with RaceID, we tested this with a pilot project focused on runners of different fitness levels, who trained for 8 weeks to run a 10km race.
Result: The results were a success! Ellida delivered personal coaching through an adaptive training program that was updated every week. Overall, the performance predictions were extremely accurate, and most runners achieved new Personal Bests (PBs).
Algorithm to Predict Injury from Retrospective Data in English Premier League
Challenge: Injuries cause players to miss games and training, which is costly in terms of money, resources, and the team losing games. Prediction algorithms are needed to detect risks early and allow for adjustment in the training programs
Solution: Most teams collect large amounts of data about each player from several sources, such as GPS-tracking, heart rate variability, physiological and mobility screenings, subjective ratings, blood markers, and injury and medical records. These data are readily available and can be used to annotate injuries even at subclinical levels
Result: A more complete player profile, better training data for injury prediction and an algorithm that can be deployed across different teams and sports
Algorithm to Predict Illness from Retrospective Data in English Premier League
Challenge: Illness is costly and categorizing them for better AI prediction algorithms is needed. Moreover, understanding the training context which could have contributed to the Illness e.g. high load, poor sleep, is needed
Solution: Multiple health and training data streams were used to predict illness at subclinical levels
Result: A more complete player profile with illness prediction that can be deployed across different teams and sports
Individual Profiles for Elite Swimmers
Challenge: All athletes respond differently to the same training, and decisions are made based on manual analysis of the athlete. There is a lack of tools to assist the coach in objectively analyzing the situation or provide decision support.
Solution: Development of individual athlete profiles with associated algorithms for improved training decision support. If our models and algorithms are correct, we should be able to also predict future performance.
Result: Our first performance prediction algorithm was tested at the 2019 FINA World Championships and missed the exact times by on average only 0.3 %! Upcoming work includes the Swedish National team using these algorithms to optimize training and performance, initially for the winter season and thereafter for the 2020 Tokyo Olympics.
ENDURANCE ATHLETE TRAINING PROFILES
Precision Profiles for Endurance Athletes Generate Training Insight
Challenge: Conventional endurance athlete training does not take advantage of individual integrated training and biological data
Solution: Multiple non-conventional data streams were integrated to better understand physiology
Result: From 2013 these insights were used to identify how different athletes tolerate tailored training intensities to maximize performance. The specific insight is that athletes with the same potential for performance can have dramatically variable response to different training regimens
Stress Management System For Esport Players
Challenge: Physiological and mental stress are key elements in performance for esport athletes and there is a lack of training solutions that address this need
Solution: Deployment of Athlete Advisor with associated sensor monitoring, analysis and algorithms
Result: Tools to monitor and predict high stress periods and subsequently recommend adjusted training
Strategic Consulting for the Swedish Women’s Handball Team
Challenge: The team lacked integrated physiological measurements
Solution: Load and recovery estimation associated with actionable training recommendations
Result: World Junior Champions 2010, 2012
New Data Driven Physiological Load Algorithm Based on Intensity Zones
Challenge: The conventional load algorithm is based on proportional increase in intensity which impacts accuracy
Solution: Previous research on individualized intensity zones was applied to the estimation of load with results validated against gold standard orthogonal methods
Result: From 2013 the more accurate and easy to deploy algorithm was applied to SVEXA Projects.
PRECISION HEART RATE INTENSITY ZONES
New Algorithm For Deciding Heart Rate Zones and Thresholds
Challenge: General heart rate zones that reflect intensity are not accurate across individuals. Training at a correct exercise intensity is crucial for appropriate training adaptations, avoiding "junk training" and minimizing the risk of overtraining and is core to physical activity guidelines
Solution: Measurable physiological and demographic parameters were used to build a model that accurately estimates individual heart rate zones that are associated with different physiological adaptations.
Result: From 2011 the individualized method of estimating heart rate zones was used to more precisely pinpoint the training effort needed to induce the appropriate training adaptations
Prediction of Overtraining in Endurance Athletes
Challenge: Overtraining in endurance athletes prevents optimal performance and increases the chance of injuries
Solution: Integrated data analysis including objective, relative, subjective and metabolomic variables were used to predict gold standard estimates of overtraining
Result: A predictive algorithm that can identify the early signs of overtraining
Predicting the Performance in Runners from Novice to Elite
Challenge: Measuring one variable, for example, VO2 max, does not give the full picture of a runner's performance level. This is standard practice
Solution: Integration of number of proprietary parameters into a model shows marked accuracy improvement
Result: Accurate prediction of running performance and subsequent training advice for future improvements
Estimation of Stride Dynamics Parameters for Optimization of Running Economy
Challenge: There is limited actionable information on stride dynamics parameters as ground contact time, stride length and vertical oscillation as related to performance
Solution: More than 200 runners, ranging in ability from world-class runners to beginners, were recruited. Running economy was determined with indirect calorimetry together with accelerometry-based assessment of stride dynamics
Result: Running economy and stride dynamics needs to be adjusted for each runners body size, performance level and physiological profile. An algorithm based on these parameters was developed
Detection of Running Surface Using Accelerometry
Challenge: Different running surfaces impact physiological parameters and maintaining a good running economy regardless of surface is a key performance indicator in successful trail- cross country runners and orienteers
Solution: Readily available data from chest straps was used to train an algorithm to detect road surface (tarmac, trail, gravel, off-trail)
Result: The algorithm is used to make more accurate energy expenditure and load estimates and can detect stride patterns associated with good running economy in different terrain types and at specific surfaces