SPAE: Lifelong disk failure prediction via end-to-end GAN-based anomaly detection with ensemble update

Yu Liu, Yunchuan Guan, Tianming Jiang, Ke Zhou, Hua Wang, Guangxing Hu, Ji Zhang, Wei Fang, Zhuo Cheng, Ping Huang

Abstract: Highlights•We pioneer the use of GAN-based anomaly detection for disk failure prediction to deal with the cold start problem.•Based on the GAN’s construction and loss functions, we pioneer the use of ensemble learning for prediction model update to deal with the aging problem.•We propose a novel data construction pattern, transforming the SMART data into 2D image-like representations and enabling the deployment of deep learning techniques with automatic feature extraction.•Conducting experiments on real-world datasets, we simulate the use of our model in both the initial period and the long-term use and validate its applicability for large- and small-sized datasets. The experimental results demonstrate the effectiveness of our model.