Threshold curve not smoooth in multi-label classification

Hi all,

While implementing multi-label classification for bear images as part of “Further research” in Chapter 6, I experienced this accu v/s threshold graph.

  • What is optimal thresh I should choose?
  • The curve isn’t smooth, so should i use any over way/ concept to get the thresh?
  • As the curve isn’t smooth, am I overfitting the validation set if I choose threshold value like this?
  • What other metrics can i graph with threshold, which can give a smoother curve with a clear optimal threshold?