Thread by @KickbackGAP

Round 2 of our #corruptionspotlight in which we highlight a paper on (anti-)corruption.
This time:
"Detecting fraud, corruption, and collusion in international development contracts: The design of a proof-of-concept automated system"
by Grace, Rai, Redmiles & Ghani in @IEEEorg

1/ Research has shown that #AI can be used to detect fraud (e.g creditcard misuse),
less is known about its potential to identify #corruption and #collusion
--> the author tackle this issue by developing a new algorithm.
Here is how they did in a nutshell 👇

2/ They use
a) openly available data on ~ 200,000 @WorldBank contracts (finances.worldbank.org/widgets/kdui-w…) and
b) data on over 4000 allegations within the World Bank.

What is the most often reason for allegations you might wonder? 🤔
A: Corruption in form of bribery (see 🔥map)

3/ Using a nifty classification pipeline to match the data sets the authors train an #AI model to predict substantiated allegations = justified reasons for (corruption) complaints

4/ How well does their model perform?
A: 70% accuracy
Folks, this is a 84% increase from the current investigation success rate!

5 & fin/
We feel this paper deserves much more attention as it can help to calculate a risk score for public tendors and thus direct resources more efficiently
What do you all think?
Do you know other comparable projects?

Openly available download:
ieeexplore.ieee.org/document/78407…