Modified Parse-Tree Based Pattern Extraction Approach for Detecting SQLIA Using Neural Network Model.

Whatever malware protection is upcoming, still the data are prone to cyber-attacks. The most threatening Structured Query Language Injection Attack (SQLIA) happens at the database layer of web applications leading to unlimited and unauthorized access to confidential information through malicious code injection. Since feature extraction accuracy significantly influences detection results, extracting the features of a query that predominantly contributes to SQL Injection (SQLI) is the most challenging task for the researchers. So, the proposed work primarily focuses on that using modified parse-tree representation. Some existing techniques used graph representation to identify characteristics of the query based on a predefined fixed list of SQL keywords. As the complete graph representation requires high time complexity for traversals due to the unnecessary links, a modified parse tree of tokens is proposed here with restricted links between operators (internal nodes) and operands (leaf nodes) of the WHERE clause. Tree siblings from the leaf nodes comprise the WHERE clause operands, where the attackers try to manipulate the conditions to be true for all the cases. A novelty of this work is identifying patterns of legitimate and injected queries from the proposed modified parse tree and applying a pattern-based neural network (NN) model for detecting attacks. The proposed approach is applied in various machine learning (ML) models and a neural network model, Multi-Layer Perceptron (MLP). With the scrupulously extracted patterns and their importance (weights) in legitimate and injected queries, the MLP model provides better results in terms of accuracy (97.85%), precision (93.8%), F1-Score (96%), and AUC (97.8%). [ABSTRACT FROM AUTHOR]