For the past few years Baseball Musings tried to help with playing Beat the Streak. The Day by Day Database keeps track of hit streak of at least five games. In addition, two programs produce top ten lists of players with a high probability of a hit.
Here are the top picks my programs produced for use in Beat the Streak. This post mostly explains the ideas behind the calculations. In addition, this post shows tests on the Neural Network (NN). This post discusses an NN that includes the ballpark. I updated the models, and the results of those tests are here.
The Log5 Method yields these top picks:
- 0.296 — Nico Hoerner batting against Trevor Williams.
- 0.293 — Luis Arraez batting against Bryce Elder.
- 0.292 — Giovanny Urshela batting against Steven Matz.
- 0.285 — Mike Trout batting against Steven Matz.
- 0.279 — Bo Bichette batting against Tanner Houck.
- 0.275 — Tyler Stephenson batting against Michael Wacha.
- 0.272 — Shohei Ohtani batting against Steven Matz.
- 0.272 — Nick Castellanos batting against Julio Urias.
- 0.270 — Austin Hays batting against Ryan Yarbrough.
- 0.266 — T.J. Friedl batting against Michael Wacha.
In 2023, Hoerner posts a .323 BA with a .361 OBP, so most of his OBP comes from hits. That really what one wants in a streak extender. On top of that, he struck out just 12 times this season in 133 PA, giving him a BABIP of .342.
Here are the NN picks:
- 0.293, 0.734 — Luis Arraez batting against Bryce Elder.
- 0.296, 0.715 — Nico Hoerner batting against Trevor Williams.
- 0.279, 0.713 — Bo Bichette batting against Tanner Houck.
- 0.292, 0.701 — Giovanny Urshela batting against Steven Matz.
- 0.285, 0.689 — Mike Trout batting against Steven Matz.
- 0.272, 0.684 — Nick Castellanos batting against Julio Urias.
- 0.262, 0.683 — Trea Turner batting against Julio Urias.
- 0.261, 0.683 — Alex Verdugo batting against Yusei Kikuchi.
- 0.262, 0.682 — Jeff McNeil batting against Michael Lorenzen.
- 0.254, 0.681 — Harold Ramirez batting against Roansy Contreras.
One and two get switched here, with Arraez coming out on top. Let’s look at the parameters for both batters (all values represent hit averages):
Parameter: [Pitcher1Y, Pitcher3Y, Batter1Y, Batter3Y, MLB, Park]
Luis Arraez: [‘0.219’, ‘0.211’, ‘0.303’, ‘0.296’, ‘0.220’, ‘0.216’]
Nico Hoerner: [‘0.236’, ‘0.246’, ‘0.274’, ‘0.271’, ‘0.220’, ‘0.226’]
In the Log5 calculation, we are comparing the batter pitcher matchup to the league average, which in this case is .220. Elder, below league average, pulls Arraez down a bit. Williams, who is well above league average, makes Hoerner a higher hit average batter than he normally would be.
The NN assigns weights that go through an activation function. The highest weights go the the Batter3Y parameter, and Arraez dominates here. The better parameters for Hoerner concerning pitcher and park aren’t enough to bring him up to Arraez’s level.
Note, also that the Log5 is for the batter against that specific pitcher, whereas the NN produces a probability for the game. It is answering the question, if this pitcher starts against this batter in this park what is the probability of the batter collecting at least one hit in the game.
You can follow the NN results on this spreadsheet. I do not guarantee results. Your best pick is going to have about a 25% chance of not getting a hit. Good luck!