Context The dataset contains S.M.A.R.T. attributes of hard drives from 2015 to 2018 on ST4000DM000 model from BackBlaze DC. The dataset was kindly preprocessed and ready to use. Content The dataset includes hard drive S.M.A.R.T. attributes along with model, serial number, date and capacity. The dataset was greatly preprocessed. First of all, the specific model was chosen due to the greatest number of falls. Also, because of too many health drives and a small amount of failured, all failured and only 10k health drives was taken from every year. Data was processed according to the following rules: 1. For failured drives was taken 120 days before failure. 2. For health drives was taken random slice of 120 days in a year. You can find more details here: https://github.com/awant/sd_failure_predictions Acknowledgements The original BackBlaze data: https://www.backblaze.com/b2/hard-drive-test-data.html. One can use this dataset in his own use, but he have to cite BackBlaze as the source and doesn't sell data. Inspiration 1. Is it possible to find which of hard drives will be broken in the near future? 2. Is it possible to predict a day when hard drive will be failured? 3. Is it possible to generalise an approach and predict failures of other models? In order for solutions to be comparable, I suggest use 2018 year as a test data and other as a train