This work proposes a secure and dynamic VM allocation strategy for multi-tenant distributed systems using the Thompson sampling approach. The method proves more effective and secure compared to epsilon-greedy and upper confidence bound methods, showing lower regret levels.,Initially, VM allocation was static, but the unpredictable nature of attacks necessitated a dynamic approach. Historical VM data was analyzed to understand attack responses, with rewards granted for unsuccessful attacks and reduced for successful ones, influencing regret levels.,The paper introduces a Multi Arm Bandit-based VM allocation policy, utilizing a Weighted Average Ensemble Learning algorithm trained on known attacks and non-attacks. This ensemble approach outperforms traditional algorithms like Logistic Regression, SVM, K Nearest Neighbors, and XGBoost.,For suspicious activity detection, a Stacked Anomaly Detector algorithm is proposed, trained on known non-attacks. This method surpasses existing techniques such as Isolation Forest and PCA-based approaches.,Overall, this paper presents an advanced solution for VM allocation policies, enhancing cloud-based system security through a combination of dynamic allocation, ensemble learning, and anomaly detection techniques.