Ensemble Gradient Boosting and Deep Neural network framework for vulnerability severity prediction in IoT cloud security

As IoT-cloud infrastructures become increasingly scalable and connected, IoT cloud security testing has become critical for smart environments. However, the diversity of IoT security data makes automated vulnerability assessment and severity prioritization challenging. This study proposes a predictive vulnerability assessment framework for CVSS-based Vulnerability Severity Score Prediction by integrating Gradient Boosting Regressor with gradient resampling, Deep Neural Networks (DNN), and TF-IDF-based text representation. The framework combines structured and unstructured vulnerability attributes from CVE records. TF-IDF converts textual vulnerability summaries into numerical representations, which are integrated with structured vulnerability characteristics for predictive modeling. The Gradient Boosting Regressor captures nonlinear relationships among structured features, while the DNN learns complex patterns across the combined feature space. Experiments were conducted using a publicly available CVE dataset from Kaggle. The proposed model achieved an MSE of 0.0349 and an R2 score of 99.12%, outperforming the evaluated baseline models. These results demonstrate that the integrated framework can support automated vulnerability severity estimation and prioritization, enabling effective risk management in IoT-cloud environments.