“Do you want to see my curveball?”
“That moves more like a slider.”
“I grip it like a curveball, though.”
As an analyst, and someone who works in pitching development, I see this almost daily. Players pick up a grip and essentially think that the grip makes the pitch and not the action it causes. Most of the time I go with whatever the pitcher calls the pitch. Sometimes it works out, sometimes it does not. It not only confuses people but completely changes the focus on where we want to locate the pitch. Don’t believe me? Here are a few examples of some confusing pitches:
That looked like a slider, right? Well, it is Nick Anderson’s curveball.
How about this fastball by Jeffrey Springs? Would you classify it as a sinker or a four-seam? Currently, it is classified as a sinker.
And finally, this slider by Ryne Harper. Without knowing that it is called a slider, would have you classified it as one?
My goal was to come up with a solution to this problem by using unsupervised machine learning. The idea is that by starting with new clusters, we would get a better idea of what a pitch does. First, I selected features based on the characteristics of a pitch. Drop and horizontal movement being the two main features. I also included the short-form vertical movement to include how spin affects a ball on the vertical axis. Short-form (spin-induced) movement is the amount of movement, in inches, the spin generates compared to a pitch with no spin-induced movement. So, pitches with backspin will generate positive values because they counteract gravity, and pitches with topspin will generate negative values.
I then gathered my data from Baseball Savant. Each data point you will see is a pitcher’s average pitch for that specific pitch type. However, I made a few transformations. First, I separated the pitches into two groups: pitches that generate glove-side movement and pitches that generate arm-side movement. When looking at the horizontal movement charts, left-handed cutters look like spin inefficient right-handed two-seam fastballs and left-handed sliders look like spin inefficient right-handed changeups. But we know that all those pitches spin differently, so they should not end up in the same group together. Following that, multiplied all the left-handers’ horizontal movements by negative 1 so that now all pitches are viewed from a right-handed pitcher’s perspective. Much like Baseball Savant does in their calculations. However, they show their charts from the batter’s perspective whereas I prefer to show charts from the pitcher’s perspective. Here is an example from Baseball Savant from last year:
Following the completion of my transformations, I used Orange Data Mining software and settled on K-Means unsupervised clustering. Furthermore, I settled on 6 clusters for each group, glove-side movement, and arm-side movement. The 6 clusters for each group were determined by silhouette scores. This gave me 12 different pitch types a pitcher could throw. Here are the current pitch classifications vs my reclassified pitch classifications:
Averages: 95.0 mph, -13.7" of drop, 7.5" horizontal movement, 16.7" vertical movement
Heaters are your Jacob deGrom fastballs. They hardly drop because of their high velocity and high amounts of spin induced vertical movement. If you like a hard fastball like me, this is the fastball for you. For those who like visuals, Caleb Ferguson’s four-seam is a perfect example of the average heater.
Averages: 93.4 mph, -24.0" of drop, 15.9" horizontal movement, 7.4" vertical movement
Faders are your very distinctive two-seams and sinkers. Not only are they the second hardest pitch, on average, but they have the most drop and glove-side horizontal movement for a fastball. Backspin is a very small component of their spin with most of their spin coming in the form of sidespin. Lance McCullers’s sinker is a good example of a fader.
Averages: 93.2 mph, -18.6" of drop, 12.4" horizontal movement, 13.1" vertical movement
The fastball is known for its almost equal horizontal and vertical spin induced movement profiles. Averaging 93.2 mph makes it only 0.1 mph higher than the MLB average for fastballs (excluding cutters). The best way to describe it is that this fastball is just your average, run-of-the-mill fastball. An example of your standard fastball is Jordan Montgomery’s Sinker.
Averages: 91.5 mph, -24.5" of drop, 4.9" horizontal movement, 16.0" vertical movement
Sneakers are interesting pitches. Even though they are slower than the generic fastball, and get more drop, they are much more elusive. That is mainly due to their ability to drop less than expected. These fastballs are very straight and have very little horizontal movement as most of their spin is backspin. Five years ago, this pitch group probably does not even exist, but now since backspin fastballs that drop less than expected are proving to be effective, these fastballs are now much more valuable. Adam Plutko has a four-seam that demonstrates this very well.
Averages: 86.1 mph, -36.2" of drop, 13.4" horizontal movement, 1.4" vertical movement
Without the inclusion of short-form vertical movement, I am very skeptical that K-Means clustering would have distinguished this type of pitch from the genetic changeup. You can see that there is some overlap between the two changeups just based on looking at their drop in the chart above. However, what separated this changeup from a generic changeup is the very small amount short-form vertical movement. This leads me to believe that most of its spin is sidespin. We will see the generic changeup has much more positive short-from vertical movement, leading me to believe it has much more backspin. A good example of a sinking change is Aaron Nola’s changeup.
Averages: 83.4 mph, -30.4" of drop, 13.6" horizontal movement, 9.4" vertical movement
In my eyes, this is your standard, run-of-the-mill changeup. The extra backspin on this pitch holds it up much more than the sinking change. By no means does this make it a bad pitch though. They just do not generate as many swings and misses as it will appear much straighter than a sinking change because they have more backspin. Martin Perez has a good example of an average changeup.
Averages: 90.1 mph, -24.5" of drop, -2.4" horizontal movement, 9.5" vertical movement
The cutter group is made up of cutters and hard thrown hybrid-sliders. The distinctive features of these cutters are that they drop like a fader, but only have -2.4 inches of spin-induced horizontal movement towards the glove-side. This pitch also averages above 90 mph, something you will not see with other pitches that break glove-side. An example of the cutter is the pitch that Dan Altavilla has classified in Baseball Savant as a slider.
Averages: 86.4 mph, -32.5" of drop, -3.6" horizontal movement, 4.7" vertical movement
The slider group contains mostly sliders as we would expect. This group is known for its similar short-form vertical and horizontal movement patterns. However, sliders do not rely on sidespin, backspin, or topspin as the main component of their spin. They rely on gyrospin (riflespin). A key feature of gyrospin is that it inherently does not induce movement until later in ball flight when the top of the ball is pushed down due to gravity causing gyrospin to turn into backspin, topspin, or sidespin. This causes sliders to break later that a pitch with much more sidespin, backspin, or topspin. An example of a slider is Jon Gray’s slider.
Averages: 83.6 mph, -42.1" of drop, -4.2" horizontal movement, -2.3" vertical movement
What distinguishes a slurve from a slider is the spin induced vertical movement. A slider has a small amount of backspin which gives it positive short-form vertical movement compared to a slurve that has a small amount of topspin giving it negative short-form vertical movement. Slurves are generally thrown like a curveball but move much more like a slider as their spin mainly consists of gyrospin as you would see in a slider. Brady Singer’s slider demonstrates the qualities of a slurve.
Averages: 79.9 mph, -42.6" of drop, -13.8" horizontal movement, 0.4" vertical movement
Sweepers are the sliders and curveballs that seem to sweep across the plate. What makes them so different from a slider is that instead of having mostly gyrospin, they have large amounts of sidespin and very little backspin or topspin, creating large amounts of horizontal movement. This makes them much more spin efficient than sliders and more like curveballs. However, we know that curveballs drop much more so they cannot be classified as a curveball. An example of the average sweeper is Taylor Rogers’s curveball.
Averages: 79.8 mph, -58.1" of drop, -7.3" horizontal movement, -14.0" vertical movement
Droppers are named for their distinctive drop as most of their movement being induced is vertical instead of horizontal. A key thing to look for when comparing a dropper to a curveball is that they are generally thrown harder even though they generate the similar amounts of drop (-58.1" to -59.5"). They can create this high amount of drop because most of their spin is topspin. Personally, this is my favorite type of breaking pitch to watch. Brandon Workman’s Knuckle-Curve is a great example of a dropper.
Averages: 76.3 mph, -59.5" of drop, -14.4" horizontal movement, -11.9" vertical movement
Like droppers, curveballs average almost 60 inches of drop, however, they generate that drop differently. While the curveball gets much less spin-induced vertical movement, it finds that extra drop in the lack of velocity. The curveball also averages almost double the horizontal movement of a dropper. Curveballs will usually have similar sidespin and topspin numbers. Because of that, curveballs are the ones that look like they hook and drop, much like Touki Toussaint’s curveball.
Notable Pitchers Reclassified
The one major reclassification that stands out to me here is the Jacob deGrom changeup being reclassified as a fader. It does make sense though. The next closest reclassification would probably be the sinking change, but his changeup metrics are much closer to a fader than a sinking change.
The first application I can think of is player development, and more specifically, pitcher development. As I alluded to earlier, all too often pitchers think that the grip makes the pitch. It does not. If you showed three different pitchers the same curveball grip, the likelihood of them throwing similar curveballs is extremely low. This now gives us a tool to look at and describe the types of pitches that they are throwing. With that same Curveball grip, we may see pitcher A throw a sweeper, pitcher B throw a dropper, and Pitcher C throw a slurve. With these individualized clusters, we can now communicate more effectively in what the pitch is doing instead of grouping all three of those pitches as a curveball.
Another application of player development I can see this taking on is comparing pitches within a pitcher’s arsenal. Many times, we see a pitcher throwing two different pitches that look almost identical. I have worked with many athletes that claim to throw a four-seam and two-seam fastball, but they move almost identical to each other so they are fundamentally throwing the same pitch two different ways. This now allows us to show them that regardless of grip, the athlete’s pitches do not move different enough to call them two different pitches. From there we can work to make adjustments on release, grip, spin direction, etc. to make two distinctive pitches.
Is this the end-all-be-all of pitch classification? No. But, I do believe this helps us move pitch classification in a positive direction. Instead of mixing up different pitch types, we can now give distinctive names to pitches based on how they move instead of on how the pitch is gripped. This new method of looking at pitches has applications on the player development side that are worth exploring as well. This type of pitch classification goes against the grain of decades of baseball knowledge, and I know that, but maybe it is time to question how we classify pitches.