Pattern Recognition AML ML model

[Maretti (Giovanni Carpentier)]

Hi is it possible to do some pattern recognition with Elasticsearch ML?

I have a dataset with financial data that looks like this:

Timestamp,From Bank,Account,To Bank,Account,Amount Received,Receiving Currency,Amount Paid,Payment Currency,Payment Format,Is Laundering
2022/09/01 00:25,070,100428660,010,800059F50,5105.92,US Dollar,5105.92,US Dollar,Cheque,0
2022/09/01 00:16,070,100428660,00220,800132390,15509630.09,US Dollar,15509630.09,US Dollar,Cheque,0
2022/09/01 00:01,070,100428660,00220,800132390,18722.79,US Dollar,18722.79,US Dollar,Credit Card,0
2022/09/01 00:03,070,100428660,00220,800132390,19780972.52,US Dollar,19780972.52,US Dollar,Cash,0
...

I also have a file that contains money laundering patterns

BEGIN LAUNDERING ATTEMPT - CYCLE:  Max 10 hops
2022/09/01 00:03,01467,8013C4030,020,80BC62F10,58702.10,Yuan,58702.10,Yuan,ACH,1
2022/09/01 02:52,020,80BC62F10,0240229,80F025640,7332.87,Swiss Franc,7332.87,Swiss Franc,ACH,1
2022/09/02 08:44,0240229,80F025640,0217,80FD27570,26443.70,Shekel,26443.70,Shekel,ACH,1
2022/09/02 12:29,0217,80FD27570,0024856,8090E8EB0,10621.24,Canadian Dollar,10621.24,Canadian Dollar,ACH,1
2022/09/03 10:20,0024856,8090E8EB0,0071,804ABCE90,637140.60,Rupee,637140.60,Rupee,ACH,1
2022/09/03 12:08,0071,804ABCE90,0213737,805494C30,621578.18,Rupee,621578.18,Rupee,ACH,1
2022/09/03 13:24,0213737,805494C30,014290,801B949C0,7222.58,Euro,7222.58,Euro,ACH,1
2022/09/04 03:24,014290,801B949C0,0010057,803DE1580,892031.21,Yen,892031.21,Yen,ACH,1
2022/09/04 09:44,0010057,803DE1580,0028628,80ACEE280,11364.12,Australian Dollar,11364.12,Australian Dollar,ACH,1
2022/09/04 15:51,0028628,80ACEE280,01467,8013C4030,7945.55,US Dollar,7945.55,US Dollar,ACH,1
END LAUNDERING ATTEMPT - CYCLE

BEGIN LAUNDERING ATTEMPT - GATHER-SCATTER:  Max 3-degree Fan-In
2022/09/01 00:04,0119,811C597B0,0048309,811C599A0,34254.65,Saudi Riyal,34254.65,Saudi Riyal,ACH,1
2022/09/01 19:27,0150240,812D22980,0048309,811C599A0,5971.98,Saudi Riyal,5971.98,Saudi Riyal,ACH,1
2022/09/04 05:06,0222,811B83280,0048309,811C599A0,50445.58,Saudi Riyal,50445.58,Saudi Riyal,ACH,1
2022/09/04 05:03,0048309,811C599A0,0048309,811C599A0,48649.42,Saudi Riyal,48649.42,Saudi Riyal,ACH,1
2022/09/04 14:59,0048309,811C599A0,0119,811C597B0,64379.45,Saudi Riyal,64379.45,Saudi Riyal,ACH,1
END LAUNDERING ATTEMPT - GATHER-SCATTER

...

Is there a way to train a ML model to recognize these patterns and apply them to the financial dataset to detect money laundering attempts?

So far I tested out the outlier detection and anomaly detection to detect large transactions but I need something more advanced than that any idea's how I could achieve that?

[richcollier]

Technically yes. The first file/data set is a data set that has been labeled (with a field "Is Laundering") and that could be used to build a classifier that looks at the combinations/values of the other fields (as features) to attempt to learn which of those combinations/values are indicative (influence) the "Is Laundering" variable).

However, your second file/data set shows a more complex relationship of multiple transactions that are indicative of Money Laundering - much more than just the values of the fields within a single transaction. This kind of detection you cannot do with a simple classifier since there could be N number of steps to the sequence of the laundering attempt, not a single transaction (unlike Credit Card fraud, for example).

You could look into possibly using some kind of sequencing detection by using something like sequence in EQL to see if you can identify certain sequences. This could prove to be tricky as I've never seen it done in the context of AML.

In asking around, I've heard that we have a Partner (Siren - http://siren-aml.com/) that is doing some great stuff in the AML space.