Backblaze-

Based on the original BackBlaze data set in 2022, we set time window, reset the labels, pre-process the data, and finally randomly delete 10%~99% of the data in order to construct a low-quality dataset.To validate the effectiveness of our hard disk failure prediction model, we conducted experiments on a dataset collected from the Nankai-Baidu Joint Lab at Nankai University. This includes four parts: NanKai-0, NanKai-10, NanKai-50, and NanKai-80, which respectively represent the addition of ASFD time series features in datasets that do not lose, lose 10%, 50%, and 80% of data.