K. J. Somaiya

Structured Query Language Injection (SQLI) is the topmost dangerous web application vulnerability in all web attacks, and this causes serious harm to the entire web system. Due to the heterogeneous nature of this attack, its detection remains a challenging problem. Researchers started using the Machine Learning (ML) based approach to mitigate this attack, but ML-based techniques heavily depend on the accuracy of feature extraction. To get more useful reduced features and improve accuracy, consider the semantic consistency and proper probability distribution of the words. The proper reduced dimensions improve the text classification process. Therefore, this paper uses a topic modeling-based Latent Dirichlet Allocation concept as a dimensionality reduction technique to acquire informative features. It helps to grab the more useful features by considering the semantic cooccurrence between the observed words from logs. This topic-modeling concept can act here as an efficient feature reduction technique and extracts the more valuable features from the most dangerous vulnerability logs. The paper explores the efficient detection of SQLI. The ECML/PKDD-2007 HTTP traffic logs experiments used supervised ML techniques and evaluated the results using accuracy matrix, performance time, and ROC curve.