AI News, Fraud prevention in peer-to-peer (P2P) transaction networks using Neural Nets: A Node Embedding approach
- On Saturday, October 13, 2018
- By Read More
Fraud prevention in peer-to-peer (P2P) transaction networks using Neural Nets: A Node Embedding approach
Since the emergence of digital banking and online shopping, it has never been easier for companies, banks, and customers to trade goods and transfer money.
For this reason, P2P platforms strive to prevent fraudulent transactions by letting users rate each other after completing a transaction using perception or trust scores.
Using recently-developed methods for representation learning on graphs, I built a classifier that discriminates between honest sellers and fraudsters in order to estimate the likelihood that a seller will commit fraud in a future transaction.
Even though anonymity allows users to keep their sensitive data private, there are serious disadvantages that relate to fraudulent transactions: Given the problems described above and the fact that users are anonymous, it would seem like we can only count on perception scores to discriminate between good sellers and fraudsters.
After all, unlike banks and other financial institutions that work on fraud prevention technologies, we don’t have any additional information about our users such as their credit scores or their financial history.
As shown below, the Node2Vec 2D projections clearly separate nodes into two regions, as would be expected from the fact that users come from two different marketplaces (Bitcoin OTC and Bitcoin Alpha) To make more complete node representations, I concatenated 6 additional node features to each Node2Vec vector.
For each transaction in the network, I created a 40-dimensional vector representing a buyer-seller pair and I fed these vectors into a Neural Net (NN) with the following layers: For the purpose of training this NN under supervised learning, I labeled the fraud transaction class as 1 and the honest transaction class as 0.
Because of the class imbalance (89% of the training examples are honest transactions or 0's), I used bootstrapping to train the NN with a different balanced subset of training examples every 10 epochs for a total of 150 epochs.
- On Saturday, February 22, 2020
Bitcoin: How to use Bitcoin ? | Digital cryptocurrency
Bitcoin is a new currency that was created in 2009 by an unknown person using the alias Satoshi Nakamoto. Transactions are made with no middle men ...
Bitcoin - The security of transaction block chains
A detailed explanation of what makes bitcoin transaction block chains secure. More free lessons at: Video ..
Hiring 1 Developer to Build a P2P Steem/USD Gateway!
Are we ready to buy and sell Steem with any fiat currency through peer to peer transactions without ever needing to leave the Steem blockchain or sign up for ...
ICO Review: Raiden Network (RDN) - Scalable Transfers for Ethereum
Raiden is an off-chain payment network built on top of the Ethereum network. Full report by Crush Crypto: ..
William Mougayar: "Blockchains: Past, Present and Future" | Talks at Google
The Business Blockchain charts new territory in advancing our understanding of the blockchain by unpacking its elements like no other before. William ...
Crypto Candy's Bitcoin and Cryptocurrency 101
You've heard it all about bitcoin. It'll make you rich. It's a bubble. It's the future of money. But what do you think about bitcoin and cryptocurrencies? To have an ...
Crash Course Blockchain Technology
Ready to learn more about the blockchain technology? The internet has a short history when compared to the time that we have had life on earth. But in this very ...
A speech by Mr.Kim Dong-oh CEO of Cube Chain at Global Issue Forum 2018.
It was time to draw many people's expectations and sympathy as a forum to view the present and future of the block chain industry. Innovative cipher block ...
#196: Digital Transformation in Retail Banking
196: Digital Transformation in Retail Banking Banking has changed dramatically in response to the evolving expectations of consumer and business customers.