Integrating physical and tactical factors in football using positional data: a systematic review

Introduction

Football can be characterized as a complex and dynamical system where the players collect ecological information to make decisions, allowing them to gather numerical and spatial advantage through the phases of play with a goal-orientation (Duarte et al., 2012a; Folgado et al., 2014a; Hewitt, Greenham & Norton, 2016). Hence, football players carry out intermittent movements to perform individual and collective tactical actions (Clemente et al., 2020; Duarte et al., 2012b; Low et al., 2019). Tracking systems have been used to compute spatiotemporal measures and assess players’ positions during training and match settings (Lames, Erdmann & Walter, 2010; Marcelino et al., 2020; Pol et al., 2020). Positional data can be captured at different frequencies by tracking systems, such as, global navigation satellite systems (GNSS) or global positioning systems (GPS) (Beato et al., 2018; Rago et al., 2020; Teixeira et al., 2021a), local radio-based local positioning (LPM) (Hoppe et al., 2018; Leser, Baca & Ogris, 2011; Ogris et al., 2012) and computerized-video or optical-based tracking systems (Beato & Jamil, 2018; Castellano, Alvarez-Pastor & Bradley, 2014; Di Salvo & Marco, 2006). The players and ball positioning can be computed by Cartesian and Euclidian coordinates (xx, yy) contextualizing the physical demands on the tactical behaviour (Carrilho et al., 2020; Clemente et al., 2013; Low et al., 2019; Memmert, Lemmink & Sampaio, 2017). However, some of the above mentioned tracking methods do not allow to gather information on the player-ball-goal position (Carrilho et al., 2020; Vidal-Codina et al., 2022), opponent-adaptive play strategy (Memmert, 2021; Ranjitha, Nathan & Joseph, 2020) and individual tactical behavior (Laakso et al., 2022; Reis & Almeida, 2020).

Furthermore, tracking systems generate a large amount and variety of data that can be used for performance analysis in football (Rein & Memmert, 2016; Rojas-Valverde et al., 2019). Notwithstanding, it is paramount implementing multidisciplinary frameworks underpinned by sports science and computer science, making use of big data methodology, new computational procedures to extract, process and analyse data that yield practical information with an impact on training and match performance (Rico-González et al., 2021). However, integrating players’ physical performance with match-related contextual factors and tactical behaviours continues to be a challenge in football science (Teixeira et al., 2022b). Moreover, the performance analysis in football needs a multidimensional approach to capture the adaptive individual and collective behaviour (Carling et al., 2014; Gonçalves et al., 2019). This multifactorial phenomenon depends on the interplay of physical, tactical and technical drivers (Bradley & Ade, 2018; Paul, Bradley & Nassis, 2015). Thus, performance analysis in football is now focused on applying the spatiotemporal parameters from an integrative perspective (Praça et al., 2022).

A growing number of reviews and meta-analyses have been published on this topic and focusing on training and match settings (Low et al., 2019; Rago et al., 2020; Teixeira et al., 2021a). Nevertheless, previous reviews have considered each performance factor independently, describing tactical behaviour independently from physical demands (Coito et al., 2022; Ometto et al., 2018). Therefore, it is important to understand the main methodological procedures to conduct an integrative analysis of physical and tactical performance in training and match in football. Also, the published studies have calculated different physical and tactical measures by tracking positional data, wherefore a procedural standardization is needed to progress towards integrative approaches (Teixeira et al., 2022a). Thus, the aim of this study was to systematically review the published articles that integrate physical and tactical variables in football using positional data.

Methods

Results

Search results and study selection

A total of 982 titles were collected on three database (WoS = 495; Pub-Med = 232 and SportDiscus = 255). After applying the selection criteria, 153 full-text articles were screened for eligibility, having 26 articles been retained for final review. Figure 1 shows PRISMA flow diagram depicting the screening procedures and search results.

Discussion

The aim of this study was to systematically review the articles that integrated physical and tactical variables using positional data in football. Physical data used to be analysed by the player’s speed. Otherwise, positional datasets were computed by spatiotemporal features such as spatial variability or regularity of the player’s movements, complex index, coordination/synchronization using intra-team and inter-team dyads, playing space.

Positional datasets allows a more ecological insight on individual physical demands, if the data interpretation considers the contextual factors and collective behaviour through a tactical analysis (Marcelino et al., 2020; Teixeira et al., 2022a; Teixeira et al., 2021c). Several authors have emphasized the need to expand the evidence produced in football on just one performance dimension (i.e., physical/physiological, technical or tactical). It is important to apply methodologies based on integrative approaches that analyse the interplay between technical factors, key tactical/performance outcomes, collective behaviour and match-related contextual drivers (Teixeira et al., 2022b). Therefore, an integrative approach was expanded in eight articles by adding psychophysiological and technical outcomes (Bradley & Ade, 2018; Paul, Bradley & Nassis, 2015). Considering the multifactorial phenomenon of performance in team sports, it is also important to consider the influence of psychological variables on the control of physical capacities, pacing behaviour, decision-making, self-regulation, and effort perception (Branquinho et al., 2020, 2021; Ferraz et al., 2022). Also, bringing together observational methodologies should be considered when positional data is to be made meaningful with skilled and technical aspects (Anguera & Hernández Mendo, 2013; Preciado et al., 2019; Sarmento et al., 2018).

The reviewed quasi-experimental studies researched training sets and match settings. Small sided-games (SSG) was the most common training task in the studies, with only three articles addressing medium-sided (MSG) and large-sided games (LSG) (Machado et al., 2020; Folgado, Gonçalves & Sampaio, 2018; Gonçalves et al., 2017; Jara et al., 2019; Nieto et al., 2022; Praça et al., 2016), respectively. Thus, SSG formats have been further explored in the literature, mainly five- and six-sided game formats. Indeed, these SSG-based formats were previous reported as useful tools to promote significant variations in the training load, and likely in the improvement of the different domains of football training (i.e., physiological, technical, and tactical dimensions) (Branquinho, Ferraz & Marques, 2021; Clemente, Afonso & Sarmento, 2021). Also, SSG and conditioned games (SSCG) are excellent ways to enhance acquisition of motor efficiency and decision-making skills (Davids et al., 2013). A research gap remains unexplored in MSG and LSG formats, as well as, its relationship with formal game formats (i.e., 7-, 8-, 9- and 11-sided formats) (Baptista et al., 2020; Coutinho et al., 2019b, 2020; Ferraz et al., 2020; Figueira et al., 2018; Gonçalves et al., 2017; Sampaio et al., 2014).

Regarding the methodological procedures, MATLAB® routines (MathWorks, Inc., Natick, MA, USA) were employed to process raw data (xx, yy) (n = 26), transforming data points into the Universal Transverse Mercator (UTM) coordinate system (Folgado et al., 2014b; Sampaio & Maçãs, 2012). The most used correction guideline to reduce tracking signal noise was a 3 Hz Butterworth low pass filter by applying non-linear logarithms using 20 windows of 3,000 points per dataset (Coutinho et al., 2019a, 2020; Figueira et al., 2018; Folgado et al., 2015, 2019; Gonçalves et al., 2017, 2018b). Other studies adopted smaller data windows such as 1,500 data sets and a sampling frequency for signal correction (i.e., 5 Hz) (Baptista et al., 2020). Nevertheless, the sampling frequency and datasets is highly dependent on the type of non-linear method to be used, and the use of higher time-series lengths can increase the consistency of the positional data (Baptista et al., 2020; Richman & Moorman, 2000; Yentes et al., 2013). Approximate entropy (ApEn) was the most noted non-linear variable for measuring the spatial movement variability/regularity (Baptista et al., 2020; Coutinho et al., 2019b, 2020; Ferraz et al., 2020; Figueira et al., 2018; Gonçalves et al., 2017; Sampaio et al., 2014). Also, stretch index can be based on the ApEn or distance, being the most reported complex index in the reviewed literature (Coutinho et al., 2019a; Olthof, Frencken & Lemmink, 2018; Praça et al., 2021). Indeed, the entropy has been extensively reported as outstanding informational parameter to describe the variability and predictability of the players’ movements (Teixeira et al., 2022b). Hilbert transform was the most frequent method, by computing the longitudinal and lateral directions through in- and anti-phase (Coutinho et al., 2019a, 2020; Fernandes et al., 2010; Figueira et al., 2018; Folgado et al., 2015, 2019; Gonçalves et al., 2017, 2018b). Total surface area or playing space (m²) can be provided through trigonometry using the smallest convex hull and/or polygonal area delimited by the peripheral players (Coutinho et al., 2017; Folgado et al., 2019; Olthof, Frencken & Lemmink, 2018). Recently, Teixeira et al. (2022b) reported that it remains to be explored correlation matrixes, clustering methods and fractality patterns. This review confirms this assertion and opens up the possibility of exploring these metrics by integrating physical demands with individual and collective behaviour.

Match analysis software was used in three studies for notational analysis to extract technical variables, specifically the LongoMatch^® software, Match Analysis Camera Systems^® (Gonçalves et al., 2018a) and Lince software^® (Canton et al., 2021; Ric et al., 2016b). Further tactical variables were selected in the retained studies using other methodological approaches (i.e., observational and notational analysis) such as metrics based on ball possession, team networks and tactical actions classifications. Technical outcomes were mainly based on individual actions and skills characterized in quantity and success (i.e., successful passes, dribbles and shots) (Coutinho et al., 2020, 2022a, 2022b, 2022c; Folgado et al., 2019; Olthof, Frencken & Lemmink, 2018; Ric et al., 2017). Data collection was based on GPS, LPM and optical-based tracking systems ranging from 5–18 Hz. The first studies were mainly based on 5 Hz GPS (SPI-Pro X II, GPS ports, Canberra, ACT, and Australia). However, the use of sampling frequency at 5 Hz must consider some limitations in the measurement of non-linear and high-intensity paths (Portas et al., 2010; Teixeira et al., 2021a). Authors should prioritise tracking devices with sampling frequencies above 10 Hz shape with an accelerometer (Gómez-Carmona et al., 2020; Rico-González et al., 2020). The latest GPS devices already recommend a sampling frequency of 10–18 Hz, specifically 10 Hz GPS units (WIMU PRO; RealTrack Systems, Almeria, Spain) (Canton et al., 2021; Machado et al., 2022), 10 Hz GPS (S5; Catapult Innovations, Melbourne, Australia) (Folgado et al., 2019; Nieto et al., 2022), 10 Hz GPS (Polar Team Pro, Kempele, Finland) (Praça et al., 2021), 15 Hz GPS (SPIPRO; GPSports, Canberra, ACT, Australia) (Coutinho et al., 2017; Praça et al., 2016; Ric et al., 2016a) and 18.18 Hz GPS (GPEXE GK; Exelio SRL, Udine, Italy) (Jara et al., 2019). LPM devices and semiautomatic video tracking system used in the surveyed studies were the Prozone^® (ProZone Holdings Ltd, Leeds, UK) (Folgado et al., 2015) and the Inmotio Object Tracking^® (BV., Amsterdam, The Netherlands) (Olthof, Frencken & Lemmink, 2018). The integration of the different tracking systems can be further explored from an integrative perspective (Buchheit et al., 2014; Linke, Link & Lames, 2018). Also, the relationship between objective (i.e., tracking systems) and subjective (i.e., observational/notational analyses) measures should be explored in future integrative approaches, in order to make the integration of technical and tactical factors more effective (Teixeira et al., 2021b). Indeed, the non-positional data can act as an added value to the positional raw data by making the information gathered from the tracking systems more feasible and comprehensive (Praça et al., 2022; Teixeira et al., 2021b).

Likewise, internal training load was collected by heart-rate-based measures and perceived exertion. Although limitations have been reported in some studies, the perceived exertion and heart-rate maintains its feasibility in elite and sub-elite football settings, it is cost-effective and straightforward to employ (Achten & Jeukendrup, 2003; Teixeira et al., 2022a; Teixeira et al., 2021b). A study used portable optical timing system to measure neuromuscular performance (i.e., countermovement jump, CMJ). Field-based tests in football become more effective when continuous control (i.e. monitoring) is integrated into the assessment (Clemente et al., 2022; Gómez-Carmona et al., 2020).

Seven studies conducted null hypothesis statistical test (NHST) and five magnitude-based inferences (MBI). Some authors used both types of analyses, besides statistical analyses such as coefficient of variation (CV) (n = 7), effect sizes (ES) and smallest worthwhile changes (SWC) (Bernards et al., 2017; Flanagan, 2013). The application of the two statistical procedures (NHST and MBI) makes it more difficult to compare results of availed studies, and future pieces of research should further analyse the potential overvaluation and bias of the study findings (Foster, Rodriguez-Marroyo & de Koning, 2017; Welsh & Knight, 2015).

The current systematic review has limitations that should be considered. Firstly, the interpretation of the studies was only qualitative, not have been done a meta-analysis. Secondly, only studies that integrated physical and tactical measures were retained for further analysis, instead of just one dimension. For this reason, other details on physical and behavioural data may be absent (Duarte et al., 2013b; Folgado et al., 2014a; Vilar et al., 2014). Although this was made clear in the inclusion and exclusion criteria, future systematic reviews should clarify which studies used both one-dimensional and integrative approach. There are topics that can be explored in the future: (i) the development of user-friendly interfaces to depict positional data, because MATLAB® routines requiring extensive training in coding and programming are used to process and display data (Math-Works, Inc., Natick, MA, USA); (ii) developing tracking and wearables devices enabling real-time feedback to increase the practical applicability and decision making of football players and coaching staff; (iii) applying advanced data analytics and big data-based procedures using artificial intelligence, machine learning and deep learning to compute automatically physical and positional data; (iv) manipulating task constraints in MSG, LSG and different game-formats can still be better exploited; (v) woman’s football is still not analysed using physical and tactical integration.

Conclusions

Based on this systematic review, physical and tactical factors can be integrated by positional data using player’s movement speed, spatial movement (and their variability, regularity or predictability), complex indexes, playing areas, intra-team and inter-team synchronization dyads. Futures research should consider applying positional data in women’s football and explore the representativeness of the MSG and LSG in youth training settings. Although positional data is being extensively applied in semi-professional and professional football, user-friendly and real-time interfaces streaming physical and tactical outcomes should be consider to enable the widespread of this technology to all.

Supplemental Information