Fraud Detection in NoSQL Database Systems using Advanced Machine Learning

Abstract

NoSQL databases such as MongoDB and Cassandra have been rapidly adopted in recent years because of their high performance, flexibility, and scalability. These databases present new security issues compared to SQL databases. NoSQL databases are vulnerable to fraud, intrusions and data breaches due to their dynamic schemas, lack of control over access and the focus on availability. This paper examines how advanced machine-learning techniques can be used to enhance fraud and intrusion detection in NoSQL databases. We examine different machine-learning algorithms, including neural networks and support vector machines. Random forests, clustering, and random forests can be used to analyze large databases activity logs in order to identify anomalous patterns of access indicative of malicious behavior. We examine how these models are trained online to detect emerging threats, and we validate the techniques using proof-of concept experiments on a prototype NoSQL based database. Our results show high accuracy for detecting injection attacks, unauthorized query, and abnormal database traffic, with low false-positive rates.