{"resourceId":"xu-university-soccer-fatigue-multisensor-2026","versions":[{"version":"external-3ac7a24ce31e17cd562696d5286abffdf536a56eded1e8739a5a27ba14c1c440","resource":{"id":"xu-university-soccer-fatigue-multisensor-2026","title":"Collegiate soccer study: multi-sensor ML beats single load markers for fatigue prediction","organization":"Guizhou Normal University (School of Physical Education)","sector":"Higher education / sport science research","geography":"China (university football/soccer team)","publishedAt":"2026-02-18","publicationDate":"2026-02-18","eventDate":null,"sourceName":"Frontiers in Physiology","sourceLabel":"Peer-reviewed research article","sourceUrl":"https://www.frontiersin.org/journals/physiology/articles/10.3389/fphys.2026.1775906/full","evidenceClass":"academic-research","outcomeClass":"mixed","topics":["data-security","operating-model","governance-procurement"],"finding":"In 48 male collegiate players monitored 12 weeks, an XGBoost model fusing GPS, heart-rate, wellness and related features predicted operationally defined fatigue with AUC 0.895 (LOSO 0.847), outperforming sRPE- or ACWR-only models; wellness, ACWR and morning HRV dominated feature importance.","sledRelevance":"Why now: fills the archive gap on measured collegiate multi-sensor fatigue modeling (not just vendor claims or injury-review caution). Newly relevant alongside continuing U.S. wearable adoption debates even though the cohort is Chinese collegiate rather than NCAA.","evidence":"Prospective design; 536 sessions + 24 matches; reported correlations (e.g., Player Load vs TRIMP r=0.81); Table 5 model metrics; SHAP/feature ranks; explicit limitations on sample, season coverage, and missing biochemical gold standard. Fatigue is a composite label (≥2 of CMJ drop, HRV drop, low wellness, high RPE), not confirmed injury.","architectureImplications":"Expect a multi-modal pipeline (GPS/IMU + HR + daily wellness + periodic CMJ) with per-athlete baselines and careful cross-validation that respects repeated measures (LOSO-style) before any alerting workflow.","governanceImplications":"Define fatigue labels and alert actions with sports medicine authority; do not equate classification AUC with permission to alter scholarships, playing time, or return-to-play without clinician ownership.","securityPrivacyImplications":"Combines location-derived load, physiology, and subjective wellness—high sensitivity; apply minimization, role-based coach views, and research-grade consent if used beyond care.","caveats":"Not U.S. NCAA; male-only; fatigue≠injury prevention proof; single institution; equipment-specific; academic calendar effects incompletely controlled.","streamIds":["college-athletics"],"roles":{"sales":"Use as evidence that multi-sensor fusion can beat single ratios in a collegiate setting—then pivot to local revalidation. Do not sell AUC 0.895 as portable. Discovery should map existing GPS/HR/wellness stack and who owns alert decisions. Unsupported: guaranteed injury reduction or NCAA-ready accuracy.","engineering":"Fit is decision-support scoring over existing athlete-monitoring feeds, with LOSO-aware validation and explainable feature views. Prerequisites: clean session IDs, morning HRV protocol adherence, and clinician-defined labels. Proof of value is local AUC/precision at agreed alert workload—not reproduction of the paper’s headline metric.","delivery":"Pilot one team’s in-season monitoring: reproduce feature set, freeze a label definition with athletic trainers, validate offline, then trial shadowed alerts. Ownership with sports performance/medicine. Risks include alert fatigue and misuse of scores in selection."},"roleDetails":{"sales":{"customerProblem":"Programs using single load ratios or raw GPS dashboards struggle to anticipate fatigue in student-athletes with dual academic-athletic constraints.","stakeholders":["Sports performance / strength staff","Athletic trainers / team physicians","Head coach","Campus research/compliance if data are reused"],"discoveryQuestions":["Which GPS, HR, wellness, and jump tools already produce daily feeds?","Who currently decides soft vs hard practice modifications?","What false-alert rate is operationally intolerable?","Are wellness answers voluntary and coach-visible today?"],"valueHypothesis":"A locally validated multi-sensor fatigue score may improve early flagging versus ACWR/sRPE alone if alert workload stays manageable—without promising Xu’s AUC on a new roster.","potentialEngagement":"Offline validation on one team’s historical season plus a shadowed-alert pilot with sports-medicine sign-off.","unsupportedClaims":["Portable AUC 0.895 to NCAA rosters","Injury-rate reduction from deploying the model","Replacement of athletic trainers’ clinical judgment"]},"engineering":{"fit":"Supervised risk scoring over wearable and wellness features; batch or near-real-time scoring after session close.","architectureIntegration":"Ingest Catapult/Polar-class exports or equivalents into a secure athletics data store; train/evaluate with subject-wise splits; expose SHAP-style reasons to clinicians.","prerequisites":["Consistent device wear compliance","Morning HRV and wellness capture protocol","Documented fatigue/injury outcome definitions","IRB or operational privacy review as applicable"],"constraints":["Chinese collegiate cohort may not match NCAA schedules or resources","Class imbalance (~15% fatigue days in study)","Missingness and exam-week effects"],"security":"Encrypt biometric and location-derived fields; limit coach-facing views to aggregated readiness rather than raw GPS traces where possible.","proofOfValue":["Local LOSO or team-holdout AUC and precision at chosen threshold","Alerts per week vs trainer capacity","Agreement rate between alerts and clinician assessment"]},"delivery":{"work":"Map data dictionary, reproduce features, validate offline, configure shadowed alerts, train staff on interpretation, and review weekly false positives.","dependencies":["Wearable vendor exports","Sports medicine labeling time","Athlete compliance with morning measures","IT secure storage"],"ownership":"Sports performance leads the model; sports medicine owns clinical action; IT owns storage access.","skillsAdoption":"Teach staff that wellness/HRV drivers can dominate—follow up on sleep/stress rather than only cutting volume.","governanceCheckpoints":["Label definition approval","Offline performance gate","Live-alert go-live review","End-of-season retention decision"],"acceptanceCriteria":["Agreed sensitivity/specificity at sustainable alert volume","No automated playing-time decisions","Documented athlete notice about monitoring purposes"],"risks":["Alert fatigue","Athletes gaming wellness scores","Overfitting to one team’s season"]}},"retrievedAt":"2026-09-15T14:13:20Z","enrichedAt":"2026-09-15T14:16:04Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Adds daily questionnaire and morning measurement burden; ensure protocols work for injured or traveling athletes and that staff capacity exists to act on flags without unpaid overtime.","procurementImplications":"If buying a vendor ‘fatigue AI,’ require subject-wise validation on the school’s data and clear separation of research vs care uses in the contract—do not accept published AUC from another country as acceptance.","operatingModelImplications":"Move from ratio dashboards to multi-input review huddles; keep humans deciding interventions; align academic calendar blackouts with monitoring interpretation.","sourceVerification":{"openedUrl":"https://www.frontiersin.org/journals/physiology/articles/10.3389/fphys.2026.1775906/full","referenceExcerpt":"The XGBoost model achieved optimal performance (AUC = 0.895), significantly outperforming single-indicator models.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}},{"version":"external-38093a9f0f4a15ab4bdb63e7ac63d2246fe1d70a9fe431192a6ff8ef383480a8","resource":{"id":"xu-university-soccer-fatigue-multisensor-2026","title":"Collegiate soccer study: multi-sensor ML beats single load markers for fatigue prediction","organization":"Guizhou Normal University / Frontiers in Physiology","sector":"Collegiate football/soccer sports science","geography":"China (university team cohort)","publishedAt":"February 18, 2026","publicationDate":"2026-02-18","eventDate":null,"sourceName":"Frontiers in Physiology","sourceLabel":"Peer-reviewed longitudinal observational study","sourceUrl":"https://www.frontiersin.org/journals/physiology/articles/10.3389/fphys.2026.1775906/full","evidenceClass":"academic-research","outcomeClass":"mixed","topics":["infrastructure","data-security","operating-model","governance-procurement"],"finding":"Xu monitored 48 male collegiate football players for 12 weeks with GPS (Catapult OptimEye S5), Polar H10 HR, daily wellness/HRV, and weekly CMJ across 536 sessions and 24 matches. Multi-sensor XGBoost predicted operationally defined fatigue with AUC 0.895 (10-fold CI 0.862–0.928) and more conservative leave-one-subject-out AUC 0.847, outperforming sRPE-only (0.687) and ACWR-only (0.724). Feature importance ranked wellness total score (18.5%), ACWR (16.2%), and morning HRV (13.8%). Match loads were ~1.7× training; academic schedules still confound recovery.","sledRelevance":"Why now: After the 2026-10-01 edition’s Huang wearable SLR (longitudinal-validation gap) and Xiao/Guo university wearable RCT, this archived collegiate multi-sensor study is newly relevant as a concrete fusion baseline showing wellness/HRV outranking ACWR—tempering single-metric vendor pitches to NCAA strength staff.","evidence":"Full Frontiers HTML reports methods, Table 5 model comparison, SHAP discussion, position-specific external load differences, and limitations (single site, n=48, no biochemical gold standard). Player Load vs TRIMP r=0.81. Ethics IRB-GZNU-2024-156.","architectureImplications":"Collegiate load platforms need fused GPS/IMU + HR + daily wellness apps with athlete-wise validation splits—not ACWR dashboards alone—and MLOps for drift when academic calendars change.","governanceImplications":"Define fatigue labels with sports medicine; keep AI as early-warning assist; disclose to athletes how wellness scores affect training decisions.","securityPrivacyImplications":"Continuous GPS/HR/wellness are sensitive biometrics; require role-based coach access, deletion on departure, and vendor bans on secondary sale.","caveats":"China university male soccer; OptimEye S5 legacy hardware; fatigue definition study-specific; transfer to U.S. NCAA women’s/multi-sport rosters unproven.","streamIds":["college-athletics"],"roles":{"sales":"Sales framing: Strength and sports-medicine buyers hear ACWR or GPS-only accuracy claims; this study shows multi-sensor ML with wellness/HRV features beats those single markers in a collegiate cohort—but still is not U.S. NCAA proof. Sell a bounded fusion pilot with athlete-wise validation and clinician override. Unsupported: guaranteed injury reduction, women’s sport transfer, or replacing ACWR governance with a black-box score.","engineering":"Engineering framing: Fit is collegiate load fusion (GPS/IMU+HR+wellness) with XGBoost-class models and LOSO-style generalization checks. Architecture: session stores, morning HRV capture, label service for fatigue criteria, monitoring for academic-calendar drift. Prerequisites: device fleet, daily survey compliance, sports-med label owners. Constraints: small n, male-only evidence, older Catapult unit. Security: biometric vaulting and coach RBAC. Proof of value: held-out athlete AUC/calibration plus alert precision under clinician review.","delivery":"Delivery framing: Instrument one team for a mesocycle, lock fatigue criteria, train staff on wellness primacy, and gate any auto-alerts. Dependencies: S&C/sports med, wearable vendor, privacy office. Skills adoption: interpret SHAP-like drivers without overreacting to single spikes. Governance checkpoints: weekly false-positive review; stop if athlete survey burden spikes. Risks: surveillance culture and inequitable position-specific errors."},"roleDetails":{"sales":{"customerProblem":"Departments buy single load metrics that miss collegiate fatigue driven by wellness and academics.","stakeholders":["Head S&C","Sports medicine","Sport coach","IT/privacy"],"discoveryQuestions":["What fatigue definition will clinicians accept?","Can athletes complete daily wellness/HRV?","Do you have athlete-wise validation capacity?"],"valueHypothesis":"Fusing wellness, HRV, and load may improve early-warning utility over ACWR alone if humans retain training-change authority.","potentialEngagement":"12-week single-team fusion monitoring pilot with LOSO-style validation report.","unsupportedClaims":["Guaranteed injury cut","NCAA women’s transfer","Autonomous load reduction"]},"engineering":{"fit":"Multi-sensor collegiate fatigue decision-support; research-to-ops with validation gates.","architectureIntegration":"GPS/HR ingest + wellness app + feature store + model service + coach dashboard with overrides.","prerequisites":["Device compliance","Label protocol","Daily survey ops","Clinician workflow"],"constraints":["n=48 evidence base","Male soccer only","Academic confounders","Legacy GPS unit in source study"],"security":"Encrypt biometrics; RBAC by role; contractual deletion on roster exit.","proofOfValue":["Athlete-wise AUC/calibration","Alert precision@k","Survey completion ≥90%"]},"delivery":{"work":"Deploy sensors, run labeling training, validate model offline, then limited advisory alerts.","dependencies":["Wearable contract","Sports med owner","Privacy review","Athlete education"],"ownership":"Sports medicine owns clinical interpretation; S&C owns training changes.","skillsAdoption":"Teach staff to combine wellness dips with load context before acting.","governanceCheckpoints":["Label lock","Validation gate","Monthly athlete feedback"],"acceptanceCriteria":["Documented LOSO/holdout metrics","Written override policy"],"risks":["Athlete distrust","Overfitting to one team","Vendor data lock-in"]}},"retrievedAt":"2026-10-02T22:18:44-05:00","enrichedAt":"2026-10-02T22:20:30-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Daily wellness/HRV adds athlete time burden; staff need ML literacy to avoid overreacting to scores.","procurementImplications":"Prefer vendors offering fusion APIs, athlete-wise validation exports, and deletion SLAs over ACWR-only SKUs.","operatingModelImplications":"Moves load management toward multi-signal early warning with sports medicine as final arbiter.","updateExplanation":"Re-opened after 2026-10-01 Huang BMC SLR and Xiao/Guo Discover RCT to extract leave-one-subject-out AUC 0.847, SHAP/wellness-first feature ranking, and explicit contrast to single-indicator ACWR/sRPE models—substantive interpretive update for collegiate buyers evaluating fusion vs single-metric load AI, preserving canonical resource id xu-university-soccer-fatigue-multisensor-2026.","sourceVerification":{"responseId":"curl-xu-fatigue-20261002T221844","openedUrl":"https://www.frontiersin.org/journals/physiology/articles/10.3389/fphys.2026.1775906/full","referenceExcerpt":"The XGBoost model achieved optimal performance (AUC = 0.895)","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}