Enhancing Database Security: A Machine Learning Approach to Anomaly Detection in NoSQL Systems

Abstract

This paper explores the integration of machine learning, specifically recurrent neural networks (RNNs), for automated anomaly detection within NoSQL databases. It addresses the challenges of maintaining data accuracy and security in these flexible and dynamic systems, where traditional methods often fall short. The proposed model features a distributed time series database, edge computing agents, a central data management backbone, and an RNN tailored for anomaly detection. Historical data is incorporated for context, and MQTT protocol ensures efficient communication. The model is evaluated using real-world data, demonstrating its potential to detect anomalies effectively. Furthermore, the paper investigates the impact of replication degree (k) on the performance and scalability of VoltDB, a NewSQL database, using the TPC-C benchmark. Results reveal that increasing k can enhance fault tolerance without significantly sacrificing throughput or latency. This work contributes to the understanding of anomaly detection in NoSQL databases and the trade-offs between consistency and performance in distributed database systems.