amrayn / planner

📅 Issue tracking and feature planning for the website.
https://amrayn.com
0 stars 0 forks source link

Provide consistent recommendations for revision #76

Closed abumq closed 1 year ago

abumq commented 1 year ago

Currently our machine learning model is not performing well for some revision cycles, i.e, the shorter the revision cycle the sensitivity needs to be high, the higher it is the sensitivity needs to be low. Here are few reasons why its happening:

  1. We have different models for different revision cycle
  2. The data is non-linear with 6 factors some going x direction other going y direction
  3. We are using raw scoring

We are going to train a new deep learning model that would provide consistent recommendations regardless of revision cycle. The scores will finally depend on percentage instead of raw prediction.