PREDICTION OF FRAUD IN BANKING DATA BY USING MACHINE LEARNING TECHNIQUES: STACKING CLASSIFIER

Abstract

The main focus of the research is on
detecting fraudulent actions in financial data using
machine learning techniques. The detection and
prevention of fraudulent transactions is of the utmost
importance in the financial industry, making this a
significant problem. The research presents
hyperparameters for class weight tweaking with the
goal of improving fraud detection. By adjusting these
parameters, the model is able to better distinguish
between real and fraudulent transactions, which
improves the system's ability to identify fraud. Three
well-known machine learning algorithms—CatBoost,
LightGBM, and XGBoost—are strategically used in
the research. The goal of combining these algorithms
is to improve the fraud detection approach as a whole
by capitalizing on their individual capabilities.In
order to optimize hyperparameters, the research
incorporates deep learning approaches. The fraud
detection system becomes more efficient and flexible
as a result of this integration, allowing it to better
detect developing fraud strategies. Using real-world
data, the project does comprehensive assessments.
According to these tests, when compared to other
approaches, the one that combines LightGBM with
XGBoost performs better across the board. This
proves that the suggested strategy outperforms the
alternatives when it comes to identifying fraudulent
actions. One of its features is a Stacking Classifier
that takes into account both RandomForest and
LightGBM classifier predictions while taking certain
parameters into account. This ensemble method
improves prediction accuracy by combining the best
features of many models; the final estimator is a
GradientBoostingClassifier.
Hyperparameter, data imbalance, machine learning,
Bayesian optimization, deep learning, ensemble
learning, and data mining are some of the index
phrases.