floodlight

framework for data-driven sports game analysis

The increasing availability of data has had a positive impact on sports science domains and application (Morgulev, 2018). In team sports, two data sources have recently come into focus: spatiotemporal positional data and manually annotated event data (Memmert & Raabe, 2018). Typically, positional data are collected from automatic global, local, or video-based tracking systems, providing players’ positions on a field as x/y-coordinates with high temporal and spatial resolution. Event data are manually collected by tagging relevant match events, such as passes, shots, or goals, with timestamps. These data are collected in professionally organized sports like football, handball, or basketball and are used within the context of game analysis to characterize and manage player loads or analyze the collective behavior of (opposing) teams (Castellano et al., 2014; Rein & Memmert, 2016).

Regardless of the growing availability of positional and event data, the technical challenges in analyzing them remain high. On the one hand, this is due to the complexity of the data, which requires specialized parsing algorithms. On the other hand, there is currently no technical standard for data formats or temporal and spatial resolution, further limiting compatibility between data providers. Additionally, even for basic routines (e.g., field rotation, smoothing of noisy data, calculation of velocities), advanced programming skills are required to perform them to scientific standards. Such skills are rarely part of a sports science curriculum, making it an insurmountable hurdle for sports scientists and practitioners looking to work with positional and event data.

The Python library floodlight was specifically designed to simplify the entry into the analysis of positional and event data in sports games. floodlight automates basic processing routines and provides a user interface for users with minimal programming knowledge. Additionally, floodlight includes comprehensive documentation with multiple tutorials, as well as a compendium discussing the technical aspects of data analysis, routines, and design decisions. For instance, parameters of specific algorithms are fully customizable, but a default value is set based on "best practice" derived from the literature.

With this, floodlight is a pioneer for publishing code for data analysis in sports games. The library provides a sustainable platform for sports scientists and computer scientists in both the academic and applied fields, facilitating interdisciplinary collaboration with the aim of establishing open-source practices in sports science.

(copy 2)

Services

Project Duration

2021 - 2024

Project funding

Publications

  • Bassek, M., Raabe, D., Banning, A., & Memmert, D. (2022). Automatic segmentation and contextualization of elite handball matches with machine learning. In World Congress of Performance Analysis of Sport & International Conference of Computer Science in Sports (pp. 103-107). Cham: Springer Nature Switzerland.
  • Bassek, M., Raabe, D., Memmert, D., & Rein, R. (2023). Analysis of Motion Characteristics and Metabolic Power in Elite Male Handball Players. Journal of Sports Science & Medicine22(2), 310 https://doi.org/10.52082/jssm.2023.310
  • Biermann, H., Wieland, F. G., Timmer, J., Memmert, D., & Phatak, A. (2022). Towards Expected Counter-Using Comprehensible Features to Predict Counterattacks. In International Workshop on Machine Learning and Data Mining for Sports Analytics (pp. 3-13). Cham: Springer Nature Switzerland.
  • Memmert, D. (Ed.) (2021). Match Analysis. Abingdon: Routledge.
  • Memmert, D., & Raabe, D. (2018). Data Analytics in Football. Positional Data Collection, Modelling and Analysis. Abingdon: Routledge
  • Raabe et al., (2022) floodlight - A high-level, data-driven sports analytics framework. Journal of Open Source Software, 7(76), 4588. https://doi.org/10.21105/joss.04588.
  • Raabe, D., Nabben, R., & Memmert, D. (2022). Graph Representations for the Analysis of Multi-Agent Spatiotemporal Sports Data. Applied Intelligence. https://doi.org/10.1007/s10489-022-03631-z

Further Information

Contact

Project leader: Prof. Dr. Daniel Memmert