Enhancing Fraud Detection Accuracy and Adaptability Through Dynamic Feature Engineering in NoSQL Databases

EasyChair Preprint 13065

Abstract

 Fraud detection systems play a pivotal role in safeguarding organizations against financial losses and reputational damage. However, the evolving nature of fraudulent activities necessitates continual innovation in detection techniques. This abstract delves into the realm of dynamic feature engineering within NoSQL database systems, aimed at enhancing the accuracy and adaptability of fraud detection models.

Traditional fraud detection systems often rely on static features, limiting their ability to capture nuanced patterns in fraudulent behavior. In contrast, dynamic feature engineering involves the generation and updating of features in real-time, enabling fraud detection models to evolve alongside emerging threats. This abstract explores various methodologies for dynamic feature engineering within the context of NoSQL databases.

One such technique is feature hashing, which involves mapping high-dimensional data into a fixed-size space, thereby reducing computational complexity while preserving essential information. Additionally, embeddings provide a powerful means of representing categorical data in a continuous vector space, facilitating the detection of intricate relationships between variables. Furthermore, automatic feature selection algorithms enable the identification of relevant features, thereby enhancing model interpretability and efficiency.

Keyphrases: Accuracy, Adaptability, Automatic Feature Selection, Dynamic Feature Engineering, Feature Hashing, NoSQL databases, Scalability, embeddings, fraud detection, real-time detection

@booklet{EasyChair:13065,
  author    = {Dylan Stilinki and Kaledio Potter},
  title     = {Enhancing Fraud Detection Accuracy and Adaptability Through Dynamic Feature Engineering in NoSQL Databases},
  howpublished = {EasyChair Preprint 13065},
  year      = {EasyChair, 2024}}