Research ML auto classification of tickets into SEI Issue Classifications

@keithcallenberg commented during my SEI Issue Classification talk:

> Call me a hammer holder looking for nails, but when I saw "classification" in your talk title I assumed you were talking about a model for predicting the label. any thoughts about what features would be important for predicting a ticket's label?

I thought a bit:

> me says:"Improves" isn't enough... "Improves performance" - tech debt. "Improves responsiveness" - might be techdebt, might be userstory depending on if the target feature has shipped already.

I should turn this into a talk.