Efficient Detection of Anomalous User Behavior in Cloud Environments Using Simulated Mouse Dynamics and a Hybrid Ensemble Model

Uqba bn Nafaa Mohammed, Zeyad Farooq Lutfi, Raed Waheed Kadhim — Tue, 08 Sep 2026 15:23:08 +0000

This study proposes a lightweight and interpretable framework for identifying suspicious behavior within a cloud-based environment through the use of simulated mouse dynamics and a voting ensemble classifier. Contrary to existing approaches that heavily focus on the use of deep learning techniques, this study utilizes a combination of three classical machine learning classifiers: XGBoost, Random Forest, and Support Vector Machine (SVM). The proposed framework utilizes a voting classifier that achieved a test accuracy of 51.1%, effectively identifying 36 anomalous user sessions. To further increase the interpretability of this framework, this study incorporates a voting classifier based on the use of Shapley Additive Explanation (SHAP), which identifies the most influential behavioral features used to make a prediction. This study shows that a combination of engineered behavioral features, anomaly detection through Isolation Forest, and ensemble voting can provide a lightweight, scalable, and deployable framework for real-time cloud security monitoring without requiring any intrusive biometric techniques.

Journal La Multiapp

Efficient Detection of Anomalous User Behavior in Cloud Environments Using Simulated Mouse Dynamics and a Hybrid Ensemble Model