How can I improve the accuracy of my models ?
rod33
New Altair Community Member
Hello,
I have this Rapid Miner assignement to do!
I have to train and find a model that bests predict the credit downgrade of firms (variable: downgrade). I have two data sets: event.train (with "downgrade" and "Rating Rank") and event.test (without "downgrade" and "Rating Rank"). The only problem is that the best predictor variable is "Rating Rank" and so the accuracy drops from 93% to 70% when I don't train my model with "Rating Rank"...
Any suggestions?
Thanks
I have this Rapid Miner assignement to do!
I have to train and find a model that bests predict the credit downgrade of firms (variable: downgrade). I have two data sets: event.train (with "downgrade" and "Rating Rank") and event.test (without "downgrade" and "Rating Rank"). The only problem is that the best predictor variable is "Rating Rank" and so the accuracy drops from 93% to 70% when I don't train my model with "Rating Rank"...
Any suggestions?
Thanks
Tagged:
0
Best Answer
-
Hi @rod33,
This video may also give a few ideas of some manual steps you can do:
https://academy.rapidminer.com/learn/video/feature-engineering-intro
Best,
Roland0
Answers
-
Hi @rod33,
There's two main approaches to improving the model performance. The first is to improve the model; you could try different models, or increase the model complexity, and even try and run a model optimization. The second approach is to try and improve the input data, and here we might use Feature Engineering. Have you had a chance to see the RapidMiner Academy, it has some excellent content on both.
Best,
Roland0 -
thanks for you answer!
I tried with 5 models and my data is cleaned with turbo prep so i don't know where to look anymore haha...
I'll try to look into it with your advices, thanks!
If you have any other suggestions i'm here
Best,0 -
Feature Engineering looks at how we can transform the variables, and also combine them (especially numerical columns). There's an option to turn it on in Auto Model, or there's operators to help with this.
Here's two videos to get you started (the first is quite long):
Best,
Roland0 -
thanks!!
I tried and applied the video to the letter but i'm only gaining 1% in accuracy and my f1 measure is still pretty low... (45%) i'll try to look for other solutions!!0 -
Hi @rod33,
This video may also give a few ideas of some manual steps you can do:
https://academy.rapidminer.com/learn/video/feature-engineering-intro
Best,
Roland0 -
i solved the issue ! thanks tho0