Abstract:
"Cyber-attacks are one of the most serious concerns facing individuals at all levels, particularly in enterprises, as they can maliciously destroy systems and steal data. The big data available on the internet motivates hackers to launch new kind of attacks. SQL injection (SQLi) is the most common attack vector accounting for over 50% of all web application attacks, nowadays not only the SQLi also NoSQL injection attacks are getting trending among the hackers due to the lack of security optimization regarding NoSQL databases.
This suggested solution detects SQL injection from web application input forms and suggests a
SQL injection validation model based on a hybrid approach that combines CNN and RNN
(Convolutional Neural Network and Recurrent Neural Network) with BERT (Bidirectional
Encoder Representations from Transformers). High-dimensional elements of SQL injection
behavior can be exploited to fix the web application vulnerability. A actual web application input
form with typical input form validation using regex, input sanitization, and the firewall approach
is used to test the recommended strategy. The results of the recommended model analysis
demonstrate that compared to earlier methods, the NLP-based model has a higher percentage of
accuracy, recall, precision, and F1 score, making it more accurate in validating assaults.
This study demonstrates that vulnerabilities in a web application can be prevented. Advanced
technologies have been used, which will assist the developer in avoiding SQL injections in a
correct and secure manner. Due to this hybrid model, there is less chance of SQL injection hacking.
As a result, SQL injection cannot be used by hackers to gain access to systems or data."
