0

I'm working on a data mining project when I need to be able to predict chances of success on a Kickstarter project funding.

I've used a kickstarter dataset which i've found on Kaggle, and i've cleaned all the noisy data, deleted irrelevant attributes and added another useful attributes.

Now I have about 320K instances and 6 attributes.

After running J48 algorithm, I'm getting 65.07% correctly classified instances and 68.7% average roc area. I have to get this performance improved but I dont know how.

It's a college project so I have specific rules: I can only change the Confidence Factor and NumMinObj of the algorithm. I've spending a lot of time trying every combination.

What can I do else? Maybe something in my dataset is problematic?enter image description here

Snir
  • 53
  • 4

1 Answers1

0

You have a lot of instances, but few attributes. If you can't add more attributes then probably you already got the best result you can have with J48 trees, and feature selection is useless. You probably have to use a more complex classification algorithm such as RandomForest.

Fabio Celli
  • 11
  • 2