A machine learning framework built on force plate data — designed to identify athletes at risk of acute knee injury before they sustain one. 3D motion capture integration in development.
34
KINEMATIC VARIABLES
0.75
TARGET AUC
6/10
PROSPECTIVE INJURIES FLAGGED
n1-magazine.vercel.app
02
N1 PERFORMANCE LABTHE PROBLEM
WHY LOAD MANAGEMENT ISN'T ENOUGH
Workload reduction doesn't change who breaks.
Load management answers one question: how much. It does not answer the more important one: how well. Volume and intensity metrics say nothing about the movement strategy an athlete uses to absorb and produce force. Two athletes can carry identical workloads and move with entirely different levels of mechanical risk.
Non-contact knee injuries occur most often during movements an athlete has executed thousands of times. The injury is not caused by unfamiliarity. It is caused by a mechanical pattern that exceeded the tissue's tolerance on that repetition. That pattern is detectable before the injury. The signal exists in the data.
"The aim of workload reduction is to restore the athlete's baseline. The aim of N1 is to raise it."
~70%
OF ACL INJURIES ARE NON-CONTACT IN NATURE
Pre-
DETECTABLE RISK EXISTS BEFORE INJURY OCCURS
03
N1 PERFORMANCE LABDATA FOUNDATION
FORCE PLATE
GROUND CONTACT DATA
N1 uses the Hawkin Dynamics force plate as the primary assessment instrument. Every trial generates a complete ground reaction force curve sampled at 1,000Hz. From that curve, N1 extracts the impulse-momentum relationships that define how an athlete manages force through each phase of movement — braking, transition, and propulsion.
Total movement impulse and impulse asymmetry
Braking and propulsive phase durations and magnitudes
Rate of force development across movement phases
Body weight normalization for cross-athlete comparison
3D MOTION CAPTURE
IN DEVELOPMENT
JOINT KINEMATICS DATA
Force plate data tells you what happened at the ground. It does not tell you which joint created it. Motion capture closes that gap. When integrated, it will track every segment of the kinematic chain in three dimensions — giving N1 a complete picture of how force is generated, transferred, and absorbed from foot to hip on every trial.
Hip active deceleration and peak flexion angles
Femoral and tibial rotation — pattern and instantaneous
Knee flexion velocity and active deceleration capacity
Ankle dorsiflexion and foot eversion patterns
HAWKIN DYNAMICS · 3D MOTION CAPTURE (IN DEVELOPMENT)N1 PERFORMANCE LAB PH
04
N1 PERFORMANCE LABVARIABLE FRAMEWORK
THE KINEMATIC FINGERPRINT
34 VARIABLES. ONE CHAIN.
No single variable predicts a knee injury. What predicts it is the specific combination of variables — and that combination is different for every athlete. N1 captures this by measuring across six anatomical zones, building a complete mechanical fingerprint per assessment.
FORCE-TIME CURVE
Ground contact impulse metrics, total movement impulse, impulse differential
HIP
SOON
Posterior chain — peak flexion, active deceleration capacity
FEMUR / TIBIA
SOON
Rotation patterns and instantaneous rotation values at key timepoints
KNEE
SOON
Flexion angle, velocity, delta, and active deceleration
ANKLE
SOON
Dorsiflexion range, velocity, and deceleration at ground contact
FOOT
SOON
Translation and eversion — rotational pattern and instantaneous values
05
N1 PERFORMANCE LABMODEL ARCHITECTURE
WHY XGBOOST
A NON-LINEAR PROBLEM REQUIRES A NON-LINEAR SOLUTION
Hip deceleration alone does not predict a knee injury. Tibial rotation alone does not predict one either. What matters is how those variables interact — and interaction effects are invisible to linear models. A linear boundary through a non-linear problem will always miss the athletes who matter most.
Gradient boosted decision trees — xgBoost — learn non-linear boundaries by construction. Each tree in the ensemble partitions the feature space differently. Together they identify the combination of mechanical factors that separates athletes who sustain acute knee injuries from those who do not. Injury datasets are structurally imbalanced — injured athletes are always a minority. N1 corrects for this using SMOTE oversampling and stratified repeated k-fold cross-validation across every model run.
WHY SENSITIVITY
OPTIMIZE FOR THE COST THAT MATTERS
Every model makes two kinds of mistakes. The question is which one costs more. Missing an at-risk athlete has consequences that flagging a healthy one does not. One ends a season. The other costs a follow-up conversation.
COST FUNCTION
✗
FALSE NEGATIVE — at-risk athlete not identified. No intervention. Injury occurs. Season ends.
✓
FALSE POSITIVE — low-risk athlete flagged. Extra assessment. One conversation. Career intact.
N1 tunes for sensitivity — the proportion of actual injury cases the model correctly flags. A high AUC with low sensitivity is a model that looks good on paper and fails in the field. Sensitivity is the number that matters when an athlete's career is on the line.
Class imbalance addressed via SMOTE oversampling · Stratified repeated k-fold cross-validation used throughout
06
N1 PERFORMANCE LABMODEL PERFORMANCE
REFERENCE ARCHITECTURE TARGET
0.75
TARGET AUC VALUE
0.61
AVERAGE SENSITIVITY TARGET
A sensitivity of 0.61 means 6 of every 10 athletes who will sustain an acute knee injury are identified before they sustain it. An AUC of 0.75 means the model correctly ranks a randomly selected injured athlete above a randomly selected healthy one 75% of the time. Both are meaningful results for a small, imbalanced dataset.
N1 is building the assessment database required to run this model on Filipino court sport athletes. Every Hawkin Dynamics session adds one more data point. The benchmarks shown here are reference targets drawn from the sports science literature — they represent what N1 is building toward, not what the current dataset has produced.
REFERENCE ARCHITECTURE — NOT YET DERIVED FROM N1'S OWN DATASET
TARGETS WILL BE UPDATED AS N1'S ASSESSMENT DATABASE GROWS
07
N1 PERFORMANCE LABRISK DRIVER ANALYSIS
INDIVIDUAL PROFILING
SAME SCORE. DIFFERENT STORY.
Two athletes can share an identical risk score and require entirely different interventions. One may be driven by foot translation under load. The other by a hip deceleration deficit that shifts stress to the knee at landing. Acting on the wrong driver is not neutral — it spends a training block without reducing risk.
N1 extracts feature importance from the model for each individual athlete after classification. The output is a ranked list of which variables are contributing most to that athlete's score. The intervention is built from that list — not from the score alone, and not from a generic prehab template.
ATHLETE A
ATHLETE B
KEY INSIGHT
Same risk score. Athlete A's risk is driven by hip posterior chain weakness and body mass loading. Athlete B's risk is driven by foot translation and hip deceleration deficit. These require different exercise prescriptions entirely.
08
N1 PERFORMANCE LABINTERVENTION PIPELINE
DETECTION IS NOT THE DESTINATION
RISK IDENTIFIED. NOW WHAT.
01
ASSESSMENT
The athlete performs sport-specific movements on the Hawkin Dynamics force plate. N1 extracts the full ground reaction force curve and impulse-momentum breakdown across multiple trials.
02
RISK PROFILE
The model classifies the athlete into a risk tier. Feature importance extraction then identifies which specific variables are driving that classification — ranked, for this athlete, from this session.
03
TARGETED PROGRAM
Exercise selection maps to the top risk drivers — not a general injury prevention template. If hip deceleration is the driver, the block targets posterior chain eccentric capacity. If foot translation is the driver, it targets ankle and foot stability mechanics.
04
REASSESSMENT
After 6 to 8 weeks, the assessment repeats. Two questions get answered: did the target variable move, and did the risk classification change. Those answers determine what the next block addresses.
09
N1 PERFORMANCE LABSTATUS
N1 IS BUILDING THIS NOW.
●
LIVE — Force plate assessments active. Hawkin Dynamics CMJ data collected per athlete, per session. Database growing.
◐
IN DEVELOPMENT — Machine learning pipeline. Feature extraction and risk classification architecture being engineered now.
○
COMING SOON — 3D Motion Capture integration. When active, all 34 kinematic variables will be captured in a single assessment session.
○
COMING SOON — N1 Injury Risk Report. Per-athlete risk classification and ranked risk driver output delivered within 48 hours of assessment.
N1 is building a movement intelligence platform for Filipino court sport athletes — one assessment at a time. Every trial makes the model smarter. Every athlete adds to the baseline.