Fraud Detection Techniques Comparison trial and error

KNIME Community Hub
1

I found workflow for comparing detecting fraud credit card transactions models.
I’m trying to run this workflow to compare multiple models in detecting fraud transactions.
Autencoder and isolation forest are not running and I don’t know why.

The goal is to compare and play around to see how much which model detects fraud transaactions best, also using different methods , for example, I learned that if the ratio between real and fraud transactions is very high , the model would get lazy and just predict all as real. so a good solution to adjust the ratio of the dataset between real and fraud to make it 10:1 or 3:1, I would like to know what are other similar tricks to keep improving the workflow, also if there are any better workflows or better adjustments to the existing workflow

image
image1334×709 73.3 KB

1 Like
2

@simonAbdelmalak you can take a look at this article and the workflows mentioned. There also is a passage about imbalanced datasets.

Medium: “KNIME — Machine Learning and Artificial Intelligence — A Collection”

From what I see in the example you posted there are already efforts to find an optimal split with some algorithms.

1 Like