Step-by-step tutorial for predicting hard drive failures in large-scale data centers using naive bayes algorithms.
During my summer internship in 2017 at Hudson River Trading (HRT), I was assigned the task of predicting hard drive failures for machines. In modern day large-scale data centers, predicting hard drive failures can be extremely useful to streamline maintenance and backup data before the data is corrupted by a HDD failure. Thanks to the research at Backblaze, a company that specializes in cloud & backup storage services, I was able to train my predictive models with large amounts of data and achieve 99.5%+ accuracy. In this iPython notebook, I use data from Backblaze that I preprocessed to include approx. a 1:10 ratio of failed to working drives to achieve 95%+ accuracy in prediction. (This ipynb achieves lower accuracy because I use less data to train and have a higher failed:working ratio in my data than my model for HRT did).