Decoding Baseball Pitcher Intentions through Recurrent Neural Networks

June 2023 - Present

In progress development of Machine Learning analysis method for baseball pitchers using a Recurrent Neural Network. Model receives pitcher video data, makes a future pitch prediction based on given data, then demonstrates how it comes to that decision with guided backpropogation and saliency maps.

Development Plans:

  • Dataset: Use pitcher video data supplied by Occidental College Baseball to create video dataset. 4 channels of data. Each video is labeled with the pitch type.
  • Dataloader (Completed): Python script written to convert video data to format processable by pytorch dataloader. Video data is loaded into dataloader as follows: [num_frames, channels, height, width]
  • Model: Model will be a Recurrent Neural Network model. The model studies previous frames of video and attemps to make predictions on the future pitch based on body movements.