Published Paper Details:

Serverless computing has emerged as a transformative cloud paradigm that offers elastic resource provisioning, event-driven execution, and pay-per-use billing. However, these characteristics have introduced a new class of economic cyber threats known as Denial-of-Wallet (DoW) attacks, where adversaries exploit automatic scaling mechanisms to generate excessive cloud expenditure while maintaining service availability. Existing detection approaches primarily focus on traffic anomalies and resource consumption patterns but often neglect explicit cost-awareness, multi-scale temporal behavior, and model interpretability. To address these limitations, this paper proposes a Cost-Aware Multi-Scale CNN-GRU Framework with Explainable Artificial Intelligence (CMS-CG-XAI) for accurate and interpretable DoW attack detection in serverless environments. The proposed framework integrates cloud telemetry data and financial indicators through a cost-aware feature engineering module that extracts invocation cost, billing growth rate, budget utilization, and resource-to-cost conversion metrics. A one-dimensional Convolutional Neural Network (CNN) is employed to learn local behavioral patterns from invocation sequences, while a Gated Recurrent Unit (GRU) network captures long-range temporal dependencies associated with attack evolution. To improve attack characterization across multiple temporal scales, Discrete Wavelet Transform (DWT) decomposition is incorporated before temporal learning. Furthermore, SHapley Additive exPlanations (SHAP) are utilized to provide interpretable predictions and identify dominant attack-driving features. Experiments are conducted using publicly available DoW datasets and serverless telemetry benchmarks under strict chronological evaluation protocols. Performance is assessed using Accuracy, Precision, Recall, F1-Score, Matthews Correlation Coefficient (MCC), and Area Under the Receiver Operating Characteristic Curve (AUC-ROC). Comparative analysis against Random Forest, XGBoost, CNN, LSTM, GRU, and CNN-LSTM baselines demonstrates the effectiveness of the proposed framework. Results indicate that integrating cost-aware analytics with multi-scale temporal learning significantly improves detection capability while maintaining operational interpretability. The proposed framework offers a practical and scalable solution for next-generation serverless security systems and contributes toward economically aware cyber defense mechanisms in cloud-native infrastructures.