Scenario-based Synthetic Dataset Generation for Mobile Money Transactions
Loading...
Date
2022
Journal Title
Journal ISSN
Volume Title
Publisher
Federated Africa and Middle East Conference on Software Engineering
Abstract
There is limited availability of mobile money transaction datasets
from Sub-Saharan Africa for research because transaction data
records are sensitive in nature and therefore raise privacy concerns.
This has in turn hindered the potential to study fraudulent patterns
in mobile money transactions so as to propose realistic mitigation
measures based on Machine Learning Approaches to the prevailing
financial fraud challenges in the region. This research presents
mobile money scenarios that should be considered in order to implement
a simulator that can harness synthetic datasets for mobile
money transactions from Sub-Saharan Africa so as to carry out
fraud detection research. These scenarios include the definition
of a mobile money ecosystem with processes used by actors such
as mobile money agents, clients, merchants and banks to interact
with each other in mobile money operations. There is also a need
for a real mobile money dataset to extract statistical information
and diverse fraudulent behaviours of actors and fraud examples
in mobile money markets. This research uses the design considerations
to examine process-driven techniques such as numerical
simulation, agent-based modeling, and data-driven techniques such
as neural networks that can be leveraged to generate synthetic
datasets for mobile money transactions. Common data generation
toolkits like PaySim, AMLSim, RetSim and ABIDES that are based
on these techniques have been examined. The design considerations
are used to design a realistic model known as MoMTSim based on
real mobile money processes and agent-based modeling techniques
that can be implemented to generate synthetic transaction datasets
for mobile money with fraud instances. This will facilitate fraud
detection research. The synthetic datasets eliminate data privacy
risks, are easy and faster to obtain, and are cheap to experiment
with. With the proposed model, different research groups can move
to the implementation stage to realise a model for synthetic data
generation for mobile money transactions from the Sub-Saharan
region.
Description
Keywords
Mobile money, Datasets, Agent-based modeling, Fraud detection, Synthetic data
Citation
Azamuke, D., Katarahweire, M., & Bainomugisha, E. (2022, June). Scenario-based Synthetic Dataset Generation for Mobile Money Transactions. In Proceedings of the Federated Africa and Middle East Conference on Software Engineering (pp. 64-72). https://doi.org/10.1145/3531056.3542774