Fraud detection feature engineering

Please join the Simons Foundation and also our generous member organizations in supporting throughout our giving project September 23-27. 100% of your contribution will fund enhancements and brand-new efforts to advantage"s international scientific neighborhood.

You watching: Fraud detection feature engineering

All fields Title Author Abstract Comments Journal recommendation ACM classification MSC classification Report number identifier DOI ORCID author ID Aid peras Full text

See more: Real Time Physics Module 1 Solutions Manual, Page Not Found

Title:Towards automated attribute design for credit card fraud detection utilizing multi-perspective HMMs
Authors:Yvan Lucas, Pierre-Edouard Portier, Léa Laporte, Liyun He-Guelton, Olivier Caelen, Michael Granitzer, Sylvie Calabretto
Download PDF Abstract: Machine finding out and also information mining techniques have actually been offered extensively inorder to detect credit card frauds. However before, many researches take into consideration crmodify cardtransactions as isolated occasions and not as a sequence of transactions. In thisstructure, we model a sequence of crmodify card transactions from three differentperspectives, namely (i) The sequence includes or doesn"t contain a fraud (ii)The sequence is derived by fixing the card-holder or the payment terminal(iii) It is a sequence of invested amount or of elapsed time between the currentand also previous transactions. Combinations of the 3 binary perspectives giveeight sets of sequences from the (training) collection of transactions. Each one ofthese sequences is modelled via a Hidden Markov Model (HMM). Each HMMassociates a likelihood to a transactivity given its sequence of previoustransactions. These likelihoods are supplied as added features in a RandomForemainder classifier for fraud detection. Our multiple perspectives HMM-basedmethod supplies automated function engineering to model tempdental correlationships soas to boost the efficiency of the classification task and also permits for anincrease in the detection of fraudulent transactions once combined with thestate of the art skilled based attribute design strategy for crmodify cardfraud detection. In expansion to previous functions, we display that this approachgoes past ebusiness transactions and gives a durable attribute engineeringover various datasets, hyperparameters and classifiers. In addition, we comparemethods to address structural absent values.
Comments: publiburned in the journal "future generation computer system systems", in the special issue: "data exploration in the internet 3.0 age"
Subjects: Machine Learning (cs.LG); Machine Learning (stat.ML)
Cite as:
(or for this version)