A Stable Model to Predict the Hard disk failure

The increase in the production of digital information every second has raised the demand for storage as its counterpart. Hard disk serves an important storage device but subjected to failure which leads to the loss of data. To improve its reliability several researches were put forward, which didn't provide a satisfactory result and had no real-time implication of their system. The purpose of this research is to propose Disk Failure Prediction (DFP) models based on Random Forest (RF), Feed Forward Neural Network (FFNN) and unsupervised K-means clustering, with a real-time Self-Monitoring System (SMS) built on a predictive model showing the most stable and reliable performance. The performance of the model were evaluation under various test cases such as providing samples of different sizes, taking timing into account and considering voting in the predicted result. The RF model performed best under this test cases with a maximum DFP rate of 99% with 0.01% False Alarm Rate (FAR) which is superior to state of the art model such as Decision Tree (DT)