Using Machine Learning to Predict Jon Lester’s Whiff Rate

Jon Lester pitched his butt off and finished 2018 with a stellar 3.32 ERA in 180 innings. But although Lester prevented the same amount of runs as many of the game’s elite starters, the big lefty’s whiff rate was his lowest as a Cub. Why?

The answer is multifaceted and cannot be reduced to a single reason, but Lester’s curveball is one explanation. You have to flip the calendar back to 2013 to find Lester’s whiff rate on the curve as low as it was in 2018. 

In order to find out why Lester’s curveball wasn’t inducing whiffs at the same rate it had over the last few years, I used a method called decision tree classification. Specifically, I considered every single Statcast metric and told my computer to find the best predictors of Lester’s curveball whiff rate.

What did the computer say are the greatest predictors of whiffs? The greatest factor for Lester’s curve was vertical release point (release_pos_z), followed by vertical movement (pfx_z). Surprisingly, spin rate wasn’t as important as other numbers. The algorithm my computer came up with accurately predicted whiffs on 67 percent of Lester’s curveballs in 2018.

The single greatest predictor for Lester’s curve was vertical release point, followed by vertical movement. This algorithm accurately predicted whiffs on 67 percent of Lester’s curveballs in 2018.

The image below illustrates the cutoffs for release point and pitch location that best predict whiffs. If Lester’s release point is lower than 5.6795 feet, the model predicts “no whiff” (i.e., 0). But if his release point is greater than 5.6795 feet and thrown with less than -0.71935 of vertical drop, the model predicts no whiff.

Following down the tree, you can determine which variables at which cutoff predict whiffs. But, above all else, the single greatest predictor is vertical release point.

Even though we focus so much on data derived from Statcast, sometimes it really is just about getting back to basics. Producing the best results could come from something as simple as repeating mechanics. When Lester threw curves with a higher release point, he had a better chance of making the batter whiff. It might just be that simple.

And maybe one reason Lester didn’t whiff as many batters in 2018 compared to 2017 was his lower release point (image below). We’re only talking about a five percent difference, but that could be the primary factor in him generating fewer swinging strikes this past season over the previous one.

It’s nearly impossible to spot differences in release point with the naked eye, but included below are GIFs comparing two different vertical release points. The first is when Lester released above the 5.6795-foot threshold and induced a whiff. The second is when Lester released below that threshold and didn’t induce a whiff.

The velocity and location are essentially the same between the two pitches, though the second curve is slightly higher in the zone. Keep in mind that this is just one example, but I wanted to show video of what could happen if Lester’s release point is lower.

Again, what appears to matter most for Lester is his vertical release point. Perhaps letting go a little higher provides more deception, helps him command better, or generates better spin. Whatever the reason, Lester’s release point is his X-factor. Get him going over the top more and maybe those whiffs will come back in 2019.

*Code available upon request.