Building a K-Medoids Clustering technique to Detect Cloud Security-Related Anomaly

Abstract

The showiest technological advancement of the twenty-first century is perhaps cloud computing. This is because, compared to other technologies in the field, it has had the fastest widespread acceptance. The proliferation of smartphones and other mobile devices with internet connectivity has been the main driver of this acceptance. The typical person can also benefit from cloud computing; it's not simply good for businesses and organizations. The increasing volume of data exchanged between businesses and cloud service providers creates risk factors for purposeful and unintentional leaks of private information to unreliable third parties. The majority of cloud service data breaches are caused by criminal activities, insider threats, malware, weak credentials, and human mistake. The most crucial security mechanisms against complex and expanding network threats are intrusion detection systems (IDSs) and intrusion prevention systems (IPSs). The two known IDS are the Signature base capable of detecting only known attack, while the anomaly detection alerts you to suspicious behavior that is both known and unknown. Issues from the main three anomaly detection techniques were challenges in chosen threshold in statistical approach, difficulties in maintenance from Data mining approach and challenges of expanding information in machine learning. Among the three techniques categorized for anomaly detection in Cloud Machine learning is the most essential strategy the detection, since it allowed machine learning and get better with time.  From the Machine learning algorithm Recursive Feature Elimination (RFE) and K-medoids Clustering will be used to build a sophisticated model that will help achieve two separate objectives. SMO will obtain the desired nature for the model, while the K-medoids was used to build the model. The Model will be implemented using the Standard benchmark datasets, KDD CUP 99 and CICIDS2017. The performance evaluation of the proposed system was measured with an accuracy of 54.4% and 84.4% CICIDS2017 and KDDCUP 99 respectively. This shows that KDDCUP 99 is still more acceptable when clustering anomalies in cloud than the proposed CICIDS2017 dataset.