Survival Analysis For Computing Systems Using A Deep Ensemble Network

摘要

Preventing failure in computing systems is a critical challenge for every Organization around the world including the European organization for Nuclear Research (CERN). Survival analysis (time-to-event analysis) is one of the renowned techniques to estimate the survival probability of the event (i.e. system failure). Survival analysis can be used to predict future failures and schedule maintenance in advance. In general, the traditional statistical approach is widely used in the literature for survival analysis. However, most of the existing methodology considers utilizing the counting method and pre-assumes the distribution time. In this research, we propose an extension approach with deep survival analysis for a computing system of the ALICE O2-FLP project by introducing a framework that uses an ensemble neural network to learn about the characteristics of failures and predict the survival probability. We also experimented with real-world data and compared the results with the state-of-theart survival methods (i.e., Weibull, LassoCox, Random Survival Forest, DeepServ, and Neural Multi-Task Logistic Regression) to evaluate the performance of the proposed framework.