Success in team sports depends on overall team performance as well as the individual contribution of players. Performance outcomes are determined by a multitude of factors; training load, preparedness and readiness levels. The primary causes of injury are planning of the training process without objective feedback in the frame of a comprehensive approach to injury prognosis and prevention. The purpose: to predict performance by analysing all components of the team’s preparation and to determine factors related to injuries in order to better predict and prevent them.
The study examined more than 20 players of a team for the duration of one season. The data set consisted of over 2000 samples with over 150 parameters containing anthropometrical, physiological, performance, injury, and training load variables. The physiological readiness of the Central Nervous System (CNS), Cardiovascular System and Energy Supply System were frequently monitored and assessed by Omegawave 4 (Finland). Classification and regression methods such as Linear Regression, Principal Component Analysis, Decision Tree, Support Vector Machine, and Bayesia Modelling were utilised to verify the usability of the data set, using RapidMiner 5.3 (Germany) and BayesiaLab 5.2 (France). Supervised Learning and Augmented Naïve Bayes (ANB) modelling were used to construct predictive models for 3 different performance concepts. 1) A new performance metric, Overall Team Performance (OTP), was constructed to represent performance on a team level. This was achieved by weighting 14 individual performance metrics based on their correlation with Game Success (GS). Normalized OTP values were classified in the following way: 0-.3 = low, .3-.6 = medium, .6-1 = high. 2) Performance was analysed in terms of GS (win) or failure (loss). 3) Performance was analysed on an individual level. The predictive models for these performance concepts used approximately 50 parameters as predictors. Model accuracy was confirmed by a Receiver Operating Characteristic (ROC) and Confusion Matrix. Supervised Learning and ANB modelling were also used for in-depth analysis of injury. To determine the predictive power of explanatory variables, Normalized Mutual Information (NMI) was used to describe the relative influence of explanatory variables on injury. Hierarchical clustering was used to identify the components of athlete preparation and their relation to injury.
The ROC for OTP prediction was 86%. The cardiac and metabolic readiness as a part of the model, were the strongest predictors. The ROC for GS prediction was 89%. The Readiness variables alone were able to predict win (77%), but not loss. Training load was the strongest predictor of GS. Regarding physiological factors, CNS readiness was the strongest predictor, followed by cardiac readiness. Individual performance metrics were accurately predicted by physiological parameters, however the strongest predictors varied amongst performance metrics.
The first model for predicting injury contained 147 parameters with a high significant prediction accuracy (ROC = 82%). This model was simplified and improved to 48 parameters where variables with p > .05 were excluded (ROC = 75%). Physiological parameters showed a higher relation to injury than anthropometrical and training parameters combined. The 20 highest explanatory variables related with injury included 14 physiological parameters (NMI = 35%), 3 anthropometrical parameters (NMI = 15%), and 3 training parameters (NMI = 11%). The Readiness cluster included physiological parameters that define short-term adaptational changes and showed a higher relation to injury than Warm Up, Load and Performance clusters.
For each of the 3 performance concepts, cardiac and CNS readiness were consistently amongst the strongest predictors. Prognosis of injury risks and injury prevention can be significantly improved by frequent monitoring of acute adaptational changes. These findings can be used for improved management of a team’s preparation process.