Enhancing Intelligent Anomaly Detection in Cloud Backend Systems through Contrastive Learning and Sensitivity Analysis

Abstract

This study investigates anomaly detection in cloud backend systems and addresses the limitations of traditional methods under high-dimensional complex data and scarce anomaly samples. A contrastive learning-based algorithm is proposed, which constructs more discriminative latent space representations through feature mapping and representation learning and achieves effective separation of normal and abnormal patterns by jointly optimizing contrastive loss and classification loss. To validate the effectiveness of the method, comparative experiments were conducted on a public dataset, and the results show that the proposed model outperforms several mainstream approaches in terms of AUC, ACC, F1-Score, and Precision. Sensitivity experiments were also performed to analyze the effects of temperature parameter, learning rate, negative sample ratio, and environmental disturbance on model performance. The results demonstrate that proper hyperparameter selection and environmental modeling not only improve overall detection performance but also enhance robustness and stability. By combining comparative experiments with sensitivity analysis, this study comprehensively verifies the effectiveness of the contrastive learning-based anomaly detection method in cloud backend scenarios and confirms its potential application value in complex system operations.