People have always been looking to understand what makes a good pitch. With advances in pitch tracking technology and computing power, we can begin to use large amounts of data to answer this question more definitively. I’ve created a model called PitchingBot which uses machine learning to try and find what makes a good pitch.
Machine learning describes a general class of algorithms that are very flexible and “learn” patterns from large amounts of data. This means I don’t have to tell PitchingBot what I think a good pitch is, but instead I can give it a load of pitches (and the results of those pitches) and it will train itself to recognize a pitch that gives good results.
I intend to investigate a couple of key questions:
Does PitchingBot reach the same conclusions as conventional wisdom about what makes a good pitch?
Naively, I would expect a good pitch to have the following qualities: high velocity, plenty of movement, and good location in the corner of the strike zone. I will look at whether these are true for PitchingBot and how the definition of a good pitch changes with the ball/strike count.
Can we meaningfully compare and evaluate pitchers using PitchingBot?
Are the pitchers who are best according to PitchingBot those who get the best results? PitchingBot isn’t very useful if it does not agree with real pitcher performance. Read the rest of this entry »