XGBoost Regression Classifier (XRC) Model for Cyber Attack Detection and Classification Using Inception V4

Abstract

Massive reliance on practical systems has resulted in several security concerns. The ability to identify anomalies is a critical safety feature enabled by anomaly diagnostic techniques. The construction of a data system faces a significant issue in cyber security. Because of the exploitation of valuable data, cybersecurity impacts the privacy of such data. Attack incidents must be examined using an appropriate analytics approach in elevating the safety level. Design of advanced analytical, conceptual model creation gives practical guidance and prioritizes threats/attacks across the network system. There is now substantial effectiveness in attack categorization, and evaluation through Convolution Neural Network (CNN) based classifiers. In light of the drawbacks of previous approaches, this research proposes an approach relying on the Deep Learning (DL) strategies for cyberattacks detection and categorization in the context of cyberspace incidents. Likewise, this article presents an XGBoost Regression Classifier (XRC) using Inception V4 to address those restrictions. XGBoost refers to Extreme Gradient Boosting, a decentralized gradient-boosted decision tree (GBDT) supervised learning framework that is robust and can be used in a decentralized context. XGBoost is a well-known machine learning technique because of its ability to produce outstanding accuracy. The concepts of both XGBoost and Regression classifiers are integrated and represented as a suggested hybridized classifier, which is implemented in Inception V4 to further train and test the model. The proposed XRC categorizes and forecasts several common types of network cyberattacks that includes Distributed Denial of Service (DDoS), Phishing, Cross-site Scripting (CS), Internet of Things (IoT). The sigmoidal function is used as a supportive activator to the hybridized classifier to lower the erroneous ratio and increase the effectiveness. Research shows that training and testing errors were substantially decreased when using XRC. In 9 out of 13 instances, over 97% of threats are detected by the XRC, and over 75% of threats are detected in its most challenging datasets.