Hi,
I'm texing up my resume, and was hoping to report on some work I did this summer at an internship.
For my summer work, (I worked at a mobile marketing firm) I developed a predictive model to quantify the unsubscribe rate that we would see following sending a message blast. In mobile marketing, subscribers can text "STOP" to unsubscribe from lists, and I looked at the different things involved in predicting these unsubscribe rates.
I framed my problem as a multi-class problem, where each class corresponded with rates of list unsubscribe rate (for example, 0 was no unsubscribes, 1 was 0-.05% of the list would unsubscribe, etc etc).
What metric should I report when describing my results for my resume? I achieved 85% accuracy with Random Forests for the multi class (91% for a two-class formulation!). But is percent accuracy a good metric?
If possible, I want a metric that anyone would be able to understand. So in this way, I am attracted to raw accuracy. So I'm unsure if something like sensitivity/specificity (does this even work with multiclass) would be appropriate
yes I used cross validation and split up my training and test sets following rigorous methodologies
tl;dr Is classification accuracy (as a percent) a good way to report results on a resume?
[link][9 comments]