Intelligent Cyber Threat Categorization in Cloud Environments Using Scalable Machine Learning Algorithms

Abstract

The most important concerns that must be taken into consideration while creating a dependable and trustworthy computer environment are security and privacy. Computing on the cloud is no different from any other paradigm. Customers give up complete command over their data management when they rely on third-party service providers for storage and data processing, which increases the risk that these suppliers won't be reliable. This paper describes a scalable machine learning system to identify intelligent cyber-threats within cloud-IoT networks based on the TON-IoT network data. This data has 211,043 labeled cases with 44 network traffic characteristics of normal and malicious activity. Preprocessing of the data was done to remove duplications, deal with missing values, encode labels, feature selection through a Random Forest classifier, feature scaling through StandardScaler and class balancing through SMOTE. The two classification models, the Decision Tree (DT) and a hybrid Soft Voting Classifier have been implemented and evaluated on the basis of accuracy (ACC), precision (PRE), recall (REC), F1-score (F1), ROC curves and confusion matrices. Empirical data show that the Voting classifier had a high level of performance with 98.61% accuracy and 0.9994 AUC with high performance compared to the DT model and other available methods on IoT security. Trustworthy cyber-threat classification in cloud-based IoT infrastructures is made possible by the results, which demonstrate the applicability, strength, and efficiency of the suggested ensemble technique.