Abstract
With the increasing numbers of Internet-connected devices, security and privacy issues are thebiggest barriers to widespread cloud systems. Securing cloud systems has become a major concern
for everyone, including consumers, businesses, and the government. While attacks on any
system may never be completely stopped, real-time detection of threats is essential for efficient
system defense. Limited research has been done on effective intrusion detection systems for
IoT (Internet of Things) environments. In this paper, the authors provide a unique intrusion
detection system that detects security anomalies in cloud networks using machine learning algorithms.
This trust-based security paradigm acts as a service, allowing for interoperability
between the many network communications protocols used in cloud systems. The proposed
intrusion detection system is tested on real network traces for proof-of-concept and simulation
for scalability using Deep Learning techniques and XGboost algorithm. The results shows
96% accuracy and proves that the suggested intrusion detection system is capable of effectively
detecting real-world intrusions.Intrusion detection is a critical component of cybersecurity, as
it helps identify and prevent unauthorized access and malicious activities within a networked
environment. Effective intrusion detection systems rely on robust models that can accurately
classify network traffic as normal or malicious. Crucial aspects of building such models are designed
by effective selection of features, which determines which attributes of network traffic are
most relevant for intrusion detection to propose adaptive trust based model. In this analysis, the
impact of feature selection on the accuracy of intrusion detection models using three datasets:
UGR-16, KDD-Cup-1999, and NSL-KDD have been explored. Feature selection is a technique
used to reduce the dimensionality of data by selecting the most informative attributes while
discarding irrelevant ones and sometimes it became unnecessary for most cases. Hence in this
work, the goal is to evaluate whether feature selection consistently improves model performance
or if its impact varies across different datasets. The result of the UGR-16, KDD-Cup-1999, and
NSL-KDD NSL-KDD dataset using CNN without feature selection have achieve accuracy of
96%, 95% and 99% respectively which are better result than .