Zepp Universal sensor - Golf
Score distribution
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Nulla euismod condimentum felis vitae efficitur. Sed vel dictum quam, at blandit leo.
The Zepp Universal sensor has several different modes. I have data for usage in two different modes, tennis and golf. So far, the analysis that's been done has been on tennis, so let's try a different sport, golf!
This data is much more complex than the tennis data we've looked at so far... The golf sensor software records 40(!) variables and outputs an overall swing score between 0 and 100.
Let's look at some visualizations....
Score vs Hand Speed
This scatterplot has a clear outlier. Let's remove it...
Where to now?
With 40 variables, it looks like we've got our work cut out for us!
The score histogram looks to have some normality to it, which is good for analysis. Score and hand speed to have some correlation. Perhaps we could proceed by narrowing down some of the factors? Let's check out another pairs plot!
Analysis complete!
Not exactly... It does seem like we could get a magnifying glass out and find some promising correlations. But, what about removing outliers like we did in the first example? Maybe we can try another approach? Perhaps ridge regression?
Ridge Regession?
That's right. We have so many factors, a ridge regression might both be a better model and help us interpret the data more. It's difficult to see the details on this chart, but the main factor in our ridge regression is BACK_SWING_TEMPO_SLOW and it is a negative factor. It's importance is ~13. The next most significant factor is UP_DOWN_SWING_GOF. Since we only know the titles we'll have to guess to what these factors mean. Let's look at the pairs plot of the top 5 columns.
What is a ridge regression
Basically, it's a linear regression which contains a penalty term. This penalty terms scales down each factor to determine relative importance. It's basically a biased linear regression, which can improve performance by reducing variance.
What does this mean?
So far, not that much. We'll have to keep working on it...
We need your consent to load the translations
We use a third-party service to translate the website content that may collect data about your activity. Please review the details in the privacy policy and accept the service to view the translations.