Research on Recognition and Feature Extraction of SQL Attack Action

SQL injection attacks pose a significant threat in network security, primarily due to their ability to leak sensitive user information, gain unauthorized system access, and manipulate databases. Addressing the challenge of sparse publicly available SQL injection datasets and inadequate feature extraction that often diminishes model accuracy, this study introduces a novel feature extraction model. This model integrates TF-IDF and Word2Vec, enhancing the simulation of real attack scenarios by standardizing SQL injection statements within a targeted server setup. It employs advanced machine learning techniques to train the model on SQL injection data, incorporating it into a convolutional neural network to optimize parameters effectively. Comparative simulation experiments demonstrate that our SQL injection vulnerability detection model, utilizing a fuzz testing approach, substantially outperforms traditional web application source code analysis methods in terms of training efficiency and classification accuracy. Notably, this model excels at identifying SQL attack vectors with a detection accuracy exceeding 90%, confirming its superior efficacy in varied levels of vulnerability assessment and its ability to deliver precise predictive outcomes.