Anomaly-aware quantum convolutional neural network for carbon-efficient job arrival rate prediction in cloud computing

Cloud computing efficiency hinges on accurately forecasting workload fluctuations, especially for job arrivals where resource allocation must adapt in real-time. In this paper, we present QAP-Net (Quantum Arrival Prediction Network), a hybrid model that combines anomaly detection with a quantum + convolutional neural network(CNN) architecture to facilitate carbon-efficient job arrival predictions. The system employs a two-stage pipeline: initially, an Isolation Forest and long short-term memory (LSTM) network identify and reduce the impact of irregular data points in time-series logs. Subsequently, a quantum and CNN-based model called Q-ConvX module uses variational quantum circuits for refined feature extraction, balancing high predictive accuracy with computational efficiency. Our approach specifically addresses key limitations in current forecasting techniques by reducing computational overhead through a quantum-classical hybrid architecture and by improving robustness in dynamic cloud environments via explicit anomaly detection and removal during preprocessing. Beyond these methodological contributions, QAP-Net seamlessly integrates into OpenStack for dynamic auto-scaling, enabling more precise VM provisioning under varying load conditions. Experimental evaluations on Azure VM (2017, 2019) and Google Cluster (2011) datasets(GCD), with aggregation intervals of 10 and 30 minutes, underscore significant resource savings. Precisely, on the Google 2011 dataset, QAP-Net attains an under-provisioning rate of 10.47 % and an over-provisioning rate of 8.93 %, outperforming established methods while also demonstrating measurable reductions in carbon emissions during training and inference. These findings highlight the promise of quantum-classical hybrid solutions in addressing operational efficacy and environmental sustainability within contemporary cloud infrastructures.