Improved Attack Classification and Reduced Misclassification in Cloud Security with Multi-Class AdaBoost Models

Objectives: A robust capability of detecting intrusions is crucial in the monitoring and detection process of cloud security endeavors against cyber-attack. With the emergence of cloud networks, new software vulnerabilities are brought into being which create openings for cybercriminals. Adapting AdaBoost with weak classifier algorithms achieves significant enhancement of the detection rate in cloud environments. The investigation uses the precision, recall, and, respectively, F1-score to measure AdaBoost’s efficiency to determine whether the model can be relied upon and is accurate for intrusion detection. Methods: AdaBoost was used to examine AWS cloud network data and identify several features of network attack. To test the model, precision, recall, f1 score and 5-fold cross validation were applied showing the stability and the accuracy of the results. Findings: AdaBoost achieved 96% overall accuracy. All identified network threats, except class 3 attacks were detected correctly with the recall being unacceptably low at 0% for class 3 attacks due to model limitations. For each of the 5 iterations of the cross-validation, the model showed perfect accuracy, that was 100% precision, recall, and F1-score for most categories. As opposed to earlier work, which had 93.4% 1 and 88% 2, this model was more accurate. Novelty: Making use of scrutinizing misclassifications, this paper exposes the elements of the model that require review and elevates a superior analysis after an attack. The real AWS data analysis improves the proposed approach leading to better security insights and more accurate cloud defense systems. Keywords: Cloud Security, Forensic Analysis, AdaBoost, Attack Detection, Intrusion Detection