Predicting failures in hard drivers based on isolation forest algorithm using sliding window

Abstract

With the surge in the development of cloud computing and big data, the scale of the data center is constantly expanding. The data missing caused by the damage of large data center disk has become a problem that cannot be ignored. Predicting the failures of the hard drivers can effectively improve the reliability and reduce the maintenance cost of the data center. The SMART data is volume and multi-dimensional, as well as the high dynamic characteristics of data center hard disk accessing, so we propose an algorithm based on isolation forest using sliding window. The experiments show that this algorithm can effectively save the time of modeling and ensure high prediction accuracy.