Embodiments of the present disclosure provide method, device and computer product for predicting disk failure. The method disclosed herein comprising: obtaining operation data of a disk, each data item of the operation data indicating values for one or more attributes of the disk at an associated time point; identifying null values for the one or more attributes from the data items of the operation data; adjusting the operation data based at least on the identifying of the null values; and processing the adjusted operation data with a machine learning model, to obtain a failure prediction on whether the disk will fail within a predetermined time period after the associated time point.