2019/01/20 by Ioana Giurgiu, Giurgiu, Ioana, Anika Schumann +1
Computer Science · #Data Stream Mining Techniques #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Machine Learning in Healthcare #Software System Performance and Reliability #Time Series Analysis and Forecasting
paper · pdf · doi:10.48550/arxiv.1901.08554
openalex publication_date 2019/01/20 · openalex created_date 2022/07/30 · openalex updated_date 2026/07/28
Given key performance indicators collected with fine granularity as time\nseries, our aim is to predict and explain failures in storage environments.\nAlthough explainable predictive modeling based on spiky telemetry data is key\nin many domains, current approaches cannot tackle this problem. Deep learning\nmethods suitable for sequence modeling and learning temporal dependencies, such\nas RNNs, are effective, but opaque from an explainability perspective. Our\napproach first extracts the anomalous spikes from time series as events and\nthen builds an RNN classifier with attention mechanisms to embed the\nirregularity and frequency of these events. A preliminary evaluation on real\nworld storage environments shows that our approach can predict failures within\na 3-day prediction window with comparable accuracy as traditional RNN-based\nclassifiers. At the same time it can explain the predictions by returning the\nkey anomalous events which led to those failure predictions.\n