Understand the true value of force plate data. Sparta Science presents the Force Plate Machine Learning™ (FPML™) for reliable decision making.
What is a Force Plate?
A Force plate or platform is a device used to measure the forces exerted on the ground by a human body (see image below).
I often describe a force plate as a high-powered bathroom scale because the device itself is relatively simple. The force plate device is instrumented with load cells that convert a force, such as pressure, into an electrical signal that can be measured and standardized. As the force applied to the load cell increases, the electrical signal changes proportionally. The quality of the load cells are important for several reasons:
-
Plane(s) of measurement (vertical force versus other planes & moments in time)
-
Frequency of data collection (measured in Hertz), the higher frequency, the more data
-
Resolution (measured in Newtons), which is the accuracy of signal
Traditional Force Plate Use
In the late 1970’s, AMTI produced the first force plate for use within clinical settings and biomechanics labs. Shortly after, companies like Bertec and Kistler began manufacturing force plate hardware as well.
These highly instrumented lab settings often combine force plates with motion capture to identify gait abnormalities. From there, lab technicians export the data into spreadsheets for analysis and publication. Due to the limitation of manpower, subjects, and collection/processing time, it is often done using rather small datasets.
In the last ten years, more commercially available products have risen for practitioners; ForceDecks and Hawkin Dynamics, to name two.
These solutions provide a lighter, portable version for data collection, which may be useful for some applications but limited in their functionality when compared to Force Plate Machine Learning™. The field of Data Science and expanded functionality (e.g., machine learning) is growing exponentially faster than many of these sensors and hardware solutions.
Force Plates and Machine Learning are Changing the Game.
Machine learning is the study of computer algorithms that improve automatically through experience, meaning, as more good data is stored, the opportunity to provide more accurate insights grows.
To provide an oversimplified example, if a force plate can gather data at a very high frequency (1000 Hertz or 100 points per second) and a subject does five jumps (~1 second each), that is 5000 data points for that screening session.
Multiply that by a team of 20 and you are quickly at 100,000 data points for the day. If you scan weekly with this small group of 20, you now have 400,000 data points for analysis every month.
To truly provide value, though, we need to pair the force plate data with key events (i.e., an injury occurring) near in time to the force screen.
Data and trials need to be aggregated across multiple organizations in order to find any significance or insights since one team will never have enough injuries in a short span of time (~200 per location) to be predictive.
But, what happens when an organization is made up of more than just one team, like an MLB organization or in the military?
Those 400,000 data points exponentially increase to billions of data points that need to be stored. Storage of this much information cannot exist in a locally installed solution (i.e. CD ROM) or within Excel and cannot be analyzed instantly by the human mind.
This much volume requires a highly sophisticated data science infrastructure, often referred to as a data lake.
Sparta Science Force Plate Machine Learning™ (FPML™)
This massive data storage requirement is for raw signal data, which can be called upon periodically to update algorithms and create new algorithms and models as more data is acquired. Such signals are referred to as time-series data.
Sparta Science is the only solution currently able to provide storage of raw force-time curves and time-series data. More important, however, is the integration of machine learning models within our software.
Unfortunately, due to the excessive speed of checkbox adoption within the force plate community, practitioners have been led to believe that solely collecting predetermined calculated variables provides enough value to be insightful right now, and in the long term.
Not to mention the incorrect assumption that calculated biomechanical variables (ie. Rate of Force) are the ground truth “raw data” source, when in fact they are not. Even something as simple as Jump Height can be calculated in different ways based on time-series data.
When calculated variables are the only stored information (as is standard with most systems) we lose depth and granularity of data which greatly decreases its value.
Without proper standardization, storage, and aggregation, running data through machine learning models will never be possible. To use the example from above, when five jumps on a force plate using our Force Plate Machine Learning™, those 5000 data points are instantly standardized, stored, aggregated, run through machine learning models and populate on your screen as an injury risk prediction before your land from your last jump.
Meanwhile, with other force plate systems, calculated variables have little value now, and next to no future value, because you cannot go back and leverage the information; it is a fixed calculation for that individual and time it was collected. Those 5000 data points have now become 5 or 6 calculated variables and the future value of that data is drastically reduced.
Sparta FPML™ Data Analysis Solution
Machine learning, leveraging a data lake (think larger, more agile database), can instantly analyze the raw signals – infinite amounts of them. Over time then, as more injuries and force-time series data is collected, machine learning can learn and make specific predictions based on the data in the data lake.
Because force plates are much more available today, more and more practitioners are getting familiar with the data and understanding calculated biomechanical variables. So if you are currently using force plates, have access to these variables and some understanding as how to interpret them, why should you care about machine learning?
Where machine learning provides unprecedented value is finding the patterns we don’t know or can’t see, and that are too complex to identify in the traditional experimental approach.
This process is better than any randomized clinical trial because it is the ultimate unbiased approach. Many labeled or unlabeled data exist but what does it all really mean – without any human emotional attachment to specific outcomes?
In short, machine learning allows us to continue to learn that we don’t know what we don’t know; Sports science and health technology is learning the same lesson.
Big tech giants are no longer competing based on hardware functionality or software features, but based on data, insights, machine learning, and prediction. The sooner the fields of sports science and health technology understand this unprecedented value, the better we can help serve our athletes, patients, and warfighters.
