Proposing to use artificial neural Networks for NoSQL attack detection

Abstract

[EN] Relationships databases have enjoyed a certain boom in software worlds until now. These days, with the rise of modern applications, unstructured data production, traditional databases do not completely meet the needs of all systems. Regarding these issues, NOSQL databases have been developed and are a good alternative. But security aspects stay behind. Injection attacks are the most serious class of web attacks that are not taken seriously in NoSQL. This paper presents a Neural Network model approach for NoSQL injection. This method attempts to use the best and most effective features to identify an injection. The features used are divided into two categories, the first one based on the content of the request, and the second one independent of the request meta parameters. In order to detect attack payloads features, we work on character level analysis to obtain malicious rate of user inputs. The results demonstrate that our model has detected more attack payloads compare with models that work black list approach in keyword level.