Two ML Methods For Outlier Detection

Suggested method to use: Isolation Forest

It can be useful to test and evaluate more than one!

Isolation Forest To Find Outliers

This is preferably done with a 'reference' set (as the training set), and another set for the test set.

On the output, it's the classifications/branches with the shortest path that are outliers.
This is since there couldn't be drawn a good correlation between the input features and it's classification (otherwise more features would be correlated well).

This also works with categorical data as well as numeric (unlike above examples).

Data Cleaning and Preparation

Check if each JE is balanced

Find Outliers in DBSCAN Clustering

'Outliers' will be classified as 'noise' while others assigned to a cluster

Check Even Amounts

Rule Based Checks