International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 08 | Aug 2024
p-ISSN: 2395-0072
www.irjet.net
Graph-based Semi-Supervised Learning for Fraud Detection in Finance Navya Krishna Alapati VISA USA, INC ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - The financial field is an area that does not
Semi-supervised learning with a graph-based approach can increase the accuracy and efficiency of fraud detection by integrating labelled and unlabeled data instead of purely supervised or unsupervised methods[2]. Highly Imbalanced Datasets - Fraud detection datasets are notoriously skewed towards the majority class, making this a challenge that semi-supervised learning grasps well. Most of those transactions or activities in these datasets are legitimate, but a tiny proportion is fraudulent[3]. This is a problem for classical machine learning models because they skew towards the dataset with more representatives and tend to underestimate (or ignore) the data minority class[4]. However, with graph-based semisupervised learning, since nodes in a graph are usually connected to their neighbors and have the same characteristics as theirs, anomalies can be detected by modeling relationships between data points that might not appear through individual information. Graph-based semisupervised learning may also help handle imbalanced datasets and generalize rapidly for evolving fraud patterns. Fraud Detection Needs to Stay Ahead -New scams are always in development, and old ones are learning how to beat current security measures, so we must actively change our defense mechanisms[5]. Additionally, since this method works by adjusting the linkages between points in its graph representation of the data that it uses to identify new credit card fraud, topology changes much more rapidly than skew or channel distributions, so a change can quickly be incorporated into the analysis, furthering making for features these are better suited[6]. In addition, graph-based semi-supervised learning can help make fraud detection in finance more efficient. For traditional methods, it becomes necessary to manually do feature engineering, selecting important features and saving them from the data to train the model. This is a slow and costly process, made even worse in cases when fraudulent activities are adaptive. Conversely, a semi-supervised learning methodology based on graphs can use the relationships among data points to automatically extract redundant feature information by itself (the labels), decreasing manual work and reducing time-to-detect fraud[7]. Credit Card Fraud Detection: As a use case of graph-based semi-supervised learning in finance, consider the example of fraud detection using credit card data. For each, we can draw a graph of financial activity with the transaction as one data point and its connection to other data points, such as how it was found in our model direction history, location information, and whatever else[8]. The model can do this by examining the relationships between these data points, enabling it to
suffer from vulnerability to various types of financial fraud, with severe losses associated with individuals and organizations. It needs to be more advanced, as using traditional rule-based systems has been proven inadequate in finding new types of fraud. As a result, there has been an increase in the demand for more sophisticated methods that can evolve with illegal activities. This paper proposes a graph-based semi-supervised learning (SSL) method for fraud detection in finance. Graph representation is a machine learning algorithm that categorizes the SSL data points into genuine and fraudulent using labelled and unlabeled behavior. Because it has more data, specifically from labelled and unlabeled samples, the SSL is trained in a larger dataset with greater diversity than the conventional method; thus, its generalization power always outperforms its traditional counterparts. Extending the graph work model to include transaction relationships and network connections is a crucial enabler, supporting complex fraud with a fast-changing nature. This makes SSL particularly well suited to detecting out-of-place behaviors that might indicate fraudulent action. To sum it up, graph-based SSL is a suitable scheme for financial fraud detection. It can retain robustness and deploy ability through the synergy of graphs with semi-supervised learning to enhance accuracy in identifying fraudulent activities. This can save financial institutions millions of dollars in losses and protect the consumers. Key Words: Vulnerability, Financial Fraud, Organization, Illegal Activities, Generalization.
1.INTRODUCTION Financial fraud is a common problem, and it continues to have the worst effects on individuals and businesses. Detection and preventing fraudulent activity are essential to maintaining confidence in the system and protecting consumers' assets. Nonetheless, fraudsters continually develop new ways to commit their crimes, and as a result, traditional methods of detecting fraudulent activity can become outdated[1]. Graph-based semi-supervised learning is a more promising method for solving the problems mentioned in finance fraud detection. Semisupervised learning is a machine-learning technique that uses graphical representations to learn patterns from unlabeled data. This uses a network of connected data points and their relationships to help find exceptions and predict potential fraud in the field of fraud detection.
© 2024, IRJET
|
Impact Factor value: 8.226
|
ISO 9001:2008 Certified Journal
|
Page 211