Short Description

The process of training is a systematic (and periodized) application of physiological and biomechanical stress. Fitness results from the body’s adjustment responses causing an increased resistance against the training stimulus. On the other hand, a downside effect of physiological strain is fatigue. The potential performance of an athlete at a certain point in time is determined by the difference of the current levels of fitness and fatigue. Systematic training load management aims at optimally balancing both effects in order to increase athletes’ potential performance in the long run.

The major challenges of training planning and work load management are: (i) both fitness and fatigue are driven by (cumulative) training load but with different impulse responses and different (nonlinear) time delays; (ii) the body’s effective (and perceived) response of a training impulse is difficult to quantify. Hence, finding the “optimal” work load is challenging since high training loads without adequate recovery may trigger unwanted adaptations and negative results, while loads with insufficient duration and intensities may not generate necessary adaptations to improve physical performance.

For optimal work load management it is necessary to differentiate between the so-called "external" load, corresponding to the (physically measured) work completed by the athlete, measured independently of his or her internal characteristics. On the other hand, the "internal" load is the individual physiological and psychological response to the external load, combined with non-sport stressors. It triggers the individual training-induced adaptations and is determined by “stress/wellbeing” factors. Since the latter vary over time and across athletes, identical external load can result into different internal load depending on the athlete’s status. In fact, there is high inter- and intra-individual variability in the physiological response to training and in the relationship between training adaptations and performance, making the “optimal” workload management a moving target.

One of the major aims of the project is a reliable quantification of athletes’ internal load in order to improve training load management on an individual basis. We will systematically record athlete-specific variables, which could serve as potential predictors/indicators of internal load. These are, on the one hand, information on training contents and intensities, athletes’ session-specific perceived load (so-called rated perceived exertion; RPE), performance metrics (e.g. speed, time, distance, power, Watt) as well as information from variables (e.g., heart rate characteristics) and and biomarkers in order to track individual strain and stress levels. Based on such athlete- and training-specific „big data“, cutting-edge concepts of statistical machine learning will be used in order to (i) quantify internal load, (ii) predict the effect of a certain training stimulus in a given athlete-specific situation, and (iii) gain a better understanding which markers may serve as reliable indicators for athlete-specific states of stress, health and recovery and how modern technology on wearables can be optimally utilized.

A further major objective is to predict the athlete- and time-specific potential performance given the individual recovery/health status and the history of (internal) training. We will introduce regular and standardized ergometer test sequences, which allow us to track the individual potential (all-out) performance through time. Via statistical modeling and machine learning we will quantify the (time-varying) relationship between load and potential performance. Combining high-dimensional and heterogeneous athlete-specific training data with data from external devices, wearables and biomarkers as well as data stemming from all-out performance tests is statistically challenging and will be the core of the project. We aim at developing alternative statistical approaches, which will evaluated against each other and may be combined. In this context we employ modern statistical approaches of machine learning, such as artificial neural networks, reinforcement learning, classification techniques, dimension reduction techniques, time series approaches as well as structural sport-scientific approaches.

The project requires expertise from statistics, computer science, mathematics, sports science and sports medicine, and builds on a close collaboration with coaches and approximately 40 top athletes from the Austrian Rowing Federation (ÖRV). The technological front-end will be an app, which (i) allows athletes to enter training data and information on individual states of wellbeing/stress/recovery, (ii) is equipped with interfaces to external devices (e.g., wearables, physical performance measurement), and (iii) provides feedback in terms of indicators, graphs, summary statistics and recommended actions (based on the underlying machine learning procedures) for athletes and coaches.