New Hybrid Model Combining NoSQL and SQL Database to Ameliorate the Performance of Big Data System

Abstract

Due to their flexibility, NoSQL databases have gained popularity as a preferred data storage destination of modern web applications and can accommodate easy growth. But they are not without risk: Technical Articles can be susceptible to injection attacks that could result in a data breach or security concern. This study explores the ability to detect such NoSQL injection attacks which involves both machine learning models and data balancing techniques. We projected several classifiers - Support Vector Machines, Decision Trees, AdaBoost, Random Forest and Logistic Regression – in data balanced with approaches such as SMOTE, Random Oversamplingand NearMiss or downsampled with either Random UnderSampling, SMOTEENN or SMOTETomek. The findings indicate that ensemble classifiers (i.e., AdaBoost and Random Forest) consistently outperform the others, achieving weighted F1-scores of 0.920--1.0 with various resampling techniques. Hybrid balancing approaches triggered a significant boost in detection malicious queries, which becomes close to the ideal performance (i.e., precision and recall) for both infrequent and frequent types of GPCs. Less complex models such as Logistic Regression and Naive Bayes also performed extremely well with hybrid resampling, attaining weighted F1-scores greater than 0.99. Discussing these results show that the hybrid resampling strategy along with ensemble learning are able to make the most robust and accurate NoSQL injection detection system to secure threats web applications without ignoring real NoSQL threats.