Tezislər / Theses

THE 3
rd
 INTERNATIONAL SCIENTIFIC CONFERENCES OF STUDENTS AND YOUNG RESEARCHERS 
dedicated to the 99
th
anniversary of the National Leader of Azerbaijan Heydar Aliyev
111
identification (UID) for each client should be found because the provided 
dataset has no user information due to the security issues, leaving it to be 
idendified. After the analysis of the columns and Exploratory Data Analysis 
(EDA), we conclude that the time difference between the last transactions, 
the number of addresses found with the payment card, the type of the card, 
billing region and the transaction amount play an integral role for the user 
identification. Various Machine and Deep Learning techniques have been 
tested and shown on the Table.
Method 
AUC 
Gradient Boosting Classifier 
0.8368 
Logistic Regression 
0.8521 
Random Forest 
0.8826
Keras Neural Network 
0.8926 
XGBoost 
0.9404 
CatBoost 
0.9408 
LightGBM 
0.9434 
XGBoost with UID
0.9471 
Ensembled model
(CatBoost & LightGBM & XGBoost with UID) 
0.9603 
As it could be seen with the suggested method with UIDs (user 
identifications for each client), the AUC score is increased to 0.9471. In order 
to have the better model, we ensembled CATBoost, LightGBM and XGBoost 
with UID method. Hence, the overall score has been increased to 0.9603 
which means the performance of the model has been improved. 
To sum up, instead of using rule-based systems with simple 
matchmaking, improved ML techniques with higher performance rate provide 
more novel ways to identify frauds. This paper conducts a comprehensive 
analysis on financial fraud detection using intelligent ways of Machine 
Learning algorithms. A set of approaches are proposed for the problem after 
completing exploratory data analysis, pre-processing and comparison of the 
algorithms. 
References
: 
[1]. Pavlo Sidelov, “All You Need to Know About Machine Learning Based Fraud Detection 
Systems”, 2021. 
[2]. S. Yusuf, and D. Ekrem, 
“
Detecting Credit Card Fraud by ANN and Logistic Regression”, 
International Symposium on Innovations in Intelligent Systems and Applications, 2011. 
[3]. Kha Shing Lim, Lam Hong Lee and Yee-Wai Sim, “A Review of Machine Learning 
Algorithms for Fraud Detection in Credit Card Transaction”, International Journal of 
Computer Science and Network Security, Vol.21 No.9, 2021. 

---