pith. sign in

arxiv: 1906.11720 · v1 · pith:5C3ZN2HInew · submitted 2019-06-27 · 📊 stat.AP · stat.OT

Detecting and classifying moments in basketball matches using sensor tracked data

Tullio Facchinetti , Rodolfo Metulini , Paola Zuccolotto This is my paper

Pith reviewed 2026-05-25 13:57 UTC · model grok-4.3

classification 📊 stat.AP stat.OT
keywords basketballsensor trackingkinematic parametersactivity detectionoffense defense classificationthreshold tuningvideo ground truth
0
0 comments X

The pith

Thresholds on players' kinematic parameters from sensor data automatically detect active periods in basketball games and classify them as offensive or defensive.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a method to automatically identify active periods during basketball games and classify them as offensive or defensive. It does so by applying thresholds to kinematic parameters derived from sensor-tracked player locations, with the thresholds tuned against a video-based ground truth in a process resembling ROC curve analysis. A reader would care because basketball games interleave active play with inactive periods, and manual video review to separate them limits the scale of data analytics for player and team performance. If successful, the approach converts raw location streams into labeled play segments that can feed directly into existing offensive and defensive evaluation tools.

Core claim

The method based on the application of thresholds to players kinematic parameters, whose values undergo a tuning strategy similar to Receiver Operating Characteristic curves, using a ground truth extracted from the video of the games, automatically identifies active periods during a game and classifies them as offensive or defensive.

What carries the argument

Threshold tuning on kinematic parameters from sensor location data, calibrated to video ground truth.

If this is right

  • Game time can be segmented into active and inactive periods without manual video inspection.
  • Active periods receive automatic offensive or defensive labels ready for further analytics.
  • Sensor-tracked location data becomes usable input for performance tools focused on offense and defense scenarios.
  • The volume of structured play data available for single-player and team evaluation increases.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same threshold approach could be tested on location data from other intermittent-action sports.
  • Real-time sensor streams might support live labeling of plays during a match.
  • Combining the segmented periods with existing offensive and defensive metrics could refine efficiency ratings.
  • Cross-league testing would show whether the tuned thresholds remain stable or require per-league recalibration.

Load-bearing premise

Kinematic parameters derived from sensor location data contain enough signal for thresholds to be tuned so they reliably reproduce video-based labels for both activity detection and offense/defense classification across games.

What would settle it

Take the tuned thresholds from one set of games and apply them to a fresh set of basketball matches with independent video ground truth; measure whether detection and classification accuracy drops sharply.

read the original abstract

Data analytics in sports is crucial to evaluate the performance of single players and the whole team. The literature proposes a number of tools for both offence and defence scenarios. Data coming from tracking location of players, in this respect, may be used to enrich the amount of useful information. In basketball, however, actions are interleaved with inactive periods. This paper describes a methodological approach to automatically identify active periods during a game and to classify them as offensive or defensive. The method is based on the application of thresholds to players kinematic parameters, whose values undergo a tuning strategy similar to Receiver Operating Characteristic curves, using a ground truth extracted from the video of the games.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper claims to describe a methodological approach for automatically detecting active periods in basketball matches from sensor-tracked player location data and classifying those periods as offensive or defensive. The method applies thresholds to kinematic parameters (derived from the tracking data); threshold values are selected via a tuning procedure analogous to ROC curve analysis, with ground truth labels obtained from video review of the games.

Significance. If the tuned thresholds prove robust and generalizable across matches, the work would be useful for sports analytics by automating the segmentation of continuous tracking streams into actionable game phases without constant manual video annotation. The explicit grounding in video-derived labels and the use of standard kinematic features from existing sensor systems are practical strengths; however, the absence of any reported performance numbers in the provided description limits assessment of whether the kinematic signal is actually sufficient for reliable reproduction of the video labels.

major comments (2)
  1. [Abstract] Abstract: The description states that the method 'automatically identifies' active periods and classifies offense/defense, yet supplies no quantitative performance metrics (accuracy, precision, recall, F1, or confusion matrices), no cross-validation or hold-out results, and no error rates. Without these, it is impossible to determine whether the ROC-style tuning actually supports the central claim or whether the procedure overfits the video ground truth used for tuning.
  2. The manuscript provides no details on how the kinematic parameters are computed from the raw sensor locations, which parameters are used, or how the ROC-like threshold selection is implemented (e.g., what is treated as the positive class, how multiple parameters are combined, or whether a single global threshold set is learned). These omissions are load-bearing because the weakest assumption identified is precisely that the sensor-derived kinematics carry enough discriminative signal; without the concrete mapping and tuning procedure, the claim cannot be evaluated.
minor comments (1)
  1. [Abstract] The abstract refers to 'a tuning strategy similar to Receiver Operating Characteristic curves' but does not specify whether true-positive/false-positive rates are computed per parameter, per player, or per team, or how the final operating point is chosen. Clarifying this would improve reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the comments, which identify opportunities to strengthen the presentation of the method. We address each major comment below and will revise the manuscript to incorporate additional details and metrics.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The description states that the method 'automatically identifies' active periods and classifies offense/defense, yet supplies no quantitative performance metrics (accuracy, precision, recall, F1, or confusion matrices), no cross-validation or hold-out results, and no error rates. Without these, it is impossible to determine whether the ROC-style tuning actually supports the central claim or whether the procedure overfits the video ground truth used for tuning.

    Authors: The manuscript presents a methodological framework whose core contribution is the threshold-based procedure itself, grounded in video labels. We agree that explicit performance numbers would allow better evaluation of how well the tuned thresholds reproduce the ground truth. In revision we will add quantitative results from the tuning process (e.g., sensitivity, specificity, and F1 at the selected operating point) together with a description of the validation approach used to guard against overfitting. revision: yes

  2. Referee: The manuscript provides no details on how the kinematic parameters are computed from the raw sensor locations, which parameters are used, or how the ROC-like threshold selection is implemented (e.g., what is treated as the positive class, how multiple parameters are combined, or whether a single global threshold set is learned). These omissions are load-bearing because the weakest assumption identified is precisely that the sensor-derived kinematics carry enough discriminative signal; without the concrete mapping and tuning procedure, the claim cannot be evaluated.

    Authors: We acknowledge that the current text does not spell out the exact computation pipeline. Kinematic parameters are derived directly from the sensor position time series as instantaneous speed (Euclidean norm of the first difference) and signed direction relative to the basket; active periods are flagged when speed exceeds a tuned threshold for a minimum duration. Offense/defense classification uses the sign of the velocity component toward the offensive basket. The ROC-style procedure treats video-labeled active intervals as the positive class, sweeps candidate thresholds, and selects the operating point that maximizes a balanced accuracy criterion; a single global threshold vector is learned per match. We will expand the Methods section with the precise formulas, parameter list, and pseudocode for the tuning step. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper describes a threshold-based method on kinematic parameters derived from sensor data, with tuning performed via an ROC-like procedure against independent video-extracted ground truth labels. This constitutes a standard supervised calibration step using external labels rather than any self-definitional loop, fitted input renamed as prediction, or load-bearing self-citation. No equations or claims in the abstract or described procedure reduce the output to an input by construction. The central claim remains falsifiable against the held-out video labels and does not invoke uniqueness theorems or ansatzes from prior self-work. This is the expected non-finding for a method that explicitly relies on external ground truth.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated. The tuning process itself implies at least one fitted threshold per kinematic variable, but values and selection criteria are not given.

pith-pipeline@v0.9.0 · 5636 in / 988 out tokens · 19932 ms · 2026-05-25T13:57:16.412237+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

21 extracted references · 21 canonical work pages

  1. [1]

    Basketball on paper: rules and tools for performance analysis

    Oliver, D. Basketball on paper: rules and tools for performance analysis. Potomac Books, Inc. (2004)

    work page 2004

  2. [2]

    Big data analytics for modeling scoring probabil- ity in basketball: The effect of shooting under high-pressure conditions

    Zuccolotto, P., Manisera, M., & Sandri, M. Big data analytics for modeling scoring probabil- ity in basketball: The effect of shooting under high-pressure conditions. International Journal of Sports Science & Coaching, vol. 13(4), pp. 569-589 (2018)

    work page 2018

  3. [3]

    Kubatko, J., Oliver, D., Pelton, K., & Rosenbaum, D. T. A starting point for analyzing bas- ketball statistics. Journal of Quantitative Analysis in Sports, vol. 3(3) (2007)

    work page 2007

  4. [4]

    J., i del Alczar, X

    Sampaio, J., McGarry, T., Calleja-Gonzlez, J., Siz, S. J., i del Alczar, X. S., & Balciunas, M. Exploring game performance in the National Basketball Association using player tracking data. PloS one, 10(7), e0132894. (2015)

    work page 2015

  5. [5]

    Basketball game- related statistics that discriminate between European players competing in the NBA and in the Euroleague

    Paulauskas, R., Masiulis, N., Vaquera, A., Figueira, B., & Sampaio, J. Basketball game- related statistics that discriminate between European players competing in the NBA and in the Euroleague. Journal of human kinetics, vol. 65, pp. 225-233 (2018)

    work page 2018

  6. [6]

    Counterpoints: Advanced defensive met- rics for nba basketball

    Franks, A., Miller, A., Bornn, L., & Goldsberry, K. Counterpoints: Advanced defensive met- rics for nba basketball. In 9th Annual MIT Sloan Sports Analytics Conference, Boston, MA. (2015)

    work page 2015

  7. [7]

    The Dwight effect: A new ensemble of interior defense analytics for the NBA

    Goldsberry, K., & Weiss, E. The Dwight effect: A new ensemble of interior defense analytics for the NBA. Sports Aptitude, LLC. Web, pp. 1-11 (2013)

    work page 2013

  8. [8]

    Modeling offensive player movement in professional basketball

    Wu, S., & Bornn, L. Modeling offensive player movement in professional basketball. The American Statistician, vol. 72(1), pp. 72-79 (2018)

    work page 2018

  9. [9]

    C., & Bornn, L

    Miller, A. C., & Bornn, L. Possession sketches: Mapping nba strategies. In MIT Sloan Sports Analytics Conference (2017)

    work page 2017

  10. [10]

    Move or die: How ball movement creates open shots in the NBA

    DAmour, A., Cervone, D., Bornn, L., & Goldsberry, K. Move or die: How ball movement creates open shots in the NBA. MIT Sloan Sports Analytics Conference (2015)

    work page 2015

  11. [11]

    and Zuccolotto, P

    Metulini, R., Marisera, M. and Zuccolotto, P. Space-Time Analysis of Movements in Basket- ball using Sensor Data. Statistics and data science: new challenges, new generations (2017)

    work page 2017

  12. [12]

    and Zuccolotto, P

    Metulini, R., Manisera, M. and Zuccolotto, P.. Sensor Analytics in Basketball. Proceedings of MathSport International 2017 Conference.( 2017)

    work page 2017

  13. [13]

    Players Movements and Team Shooting Performance: a Data Mining approach for Basketball

    Metulini, R. Players Movements and Team Shooting Performance: a Data Mining approach for Basketball. Book of short papers SIS2018, ISBN-9788891910233. Publisher: Pearson, Editors: Antonino Abbruzzo, Eugenio Brentari, Marcello Chiodi e Davide Piacentino (2018)

    work page 2018

  14. [14]

    Modelling the dynamic pattern of surface area in basketball and its effects on team performance

    Metulini, R., Manisera, M., Zuccolotto, P. Modelling the dynamic pattern of surface area in basketball and its effects on team performance. Journal of Quantitative Analysis in Sports, vol. 14.3, pp. 117-130 (2018)

    work page 2018

  15. [15]

    Filtering procedures for sensor data in basketball

    Metulini, R. Filtering procedures for sensor data in basketball. Statistics & Applicazions, vol. 15(2), pp. 133-150 (2017)

    work page 2017

  16. [16]

    H., McClish, D

    Zhou, X. H., McClish, D. K., & Obuchowski, N. A. Statistical methods in diagnostic medicine (vol. 569). John Wiley & Sons (2009)

    work page 2009

  17. [17]

    Pepe, M. S. The statistical evaluation of medical tests for classification and prediction. Medicine (2003)

    work page 2003

  18. [18]

    J., & Hand, D

    Krzanowski, W. J., & Hand, D. J. ROC curves for continuous data. Chapman and Hall/CRC (2009)

    work page 2009

  19. [19]

    Youden, W. J. Index for rating diagnostic tests. Cancer, vol. 3(1), pp. 32-35 (1950)

    work page 1950

  20. [20]

    Estimation of the Youden Index and its associated cutoff point

    Fluss, R., Faraggi, D., & Reiser, B. Estimation of the Youden Index and its associated cutoff point. Biometrical Journal: Journal of Mathematical Methods in Biosciences, vol. 47(4), pp. 458-472 (2005)

    work page 2005

  21. [21]

    Classification accuracy and cut point selection

    Liu, X. Classification accuracy and cut point selection. Statistics in medicine, vol. 31(23), pp. 2676-2686 (2012)

    work page 2012