Before commencing any type of data analysis task (be it descriptive, predictive or causal inference) (1) complex phenomena of training needs to be represented in a format that is appropriate for a given analysis and questions at hand. This process involves numerous simpli cations and assumptions, which become part of the statistical model itself. (2) When it comes to injury predictions, practitioners are interested in predicting over-use soft-tissue injuries (i.e. hamstring, quads and groin pulls) that results from training load. Training load is usually monitored and represented using GPS (i.e. Total Distance, High Speed Distance, etc), heart rate (i.e. TRIMP score, time over 90% HRmax, etc) and subjective ratings (i.e. session RPE). More complex models involves moderation of external (GPS) and internal (heart rate and session RPE) training load e ects on injury likelihood by athlete readiness (i.e. wellness questionnaire, CMJ analysis, grip strength, etc) and athlete characteristics (i.e. age, height, previous injury, etc). (2) Currently there is no consensus on how these complex data and relationships should be represented for injury prediction tasks. This technical note aims to explain one particular approach of data representation and preparation for injury prediction tasks. Generated sample data in this technical note involve season long day-to-day collection of [1] session RPE, [2] Total Distance and [3] High Speed Distance for three athletes who su ered over-use soft tissue injuries. The accompanying video details data preparation and features engineering (creating new variables from existing ones) in R-Studio (3) (which is IDE for R language) (4). R packages used in this technical note are plyr (5), dplyr (6), reshape2 (7), ggplot2 (8), TTR (9) and zoo (10). The following techniques are explained in the accompanying video: . Data representation using day-to-day approach . Exponential Moving Averages . Acute to Chronic Workload Ratio (ACWR) . Injury Lead Tags . Lag variables After preparing the data and engineering new features using the approach explained in the accompanying video, practitioners can proceed with prediction tasks using multitude of classi cation methods (11). Expanding on these techniques is beyond the scope of this technical note and interested readers are directed to provided references. Accompanying video https://vimeo.com/279226723/0299 1237
© Copyright 2018 Sport Performance & Science Reports. Sport Performance & Science Reports. Alle Rechte vorbehalten.