Predictive maintenance

Early warning system for possible failures of mass storage devices based on machine learning mechanisms.

The script uses a RandomForest classifier, which based on the appropriate Self-Monitoring, Analysis and Reporting Technology (S.M.A.R.T.) parameters of the drive is able to determine the possibility of failure. The prediction was tested mostly on data from HDDs for 3, 7 and 14 days before failure. The prediction accuracy is approximately 96.4%-100%. More information in docs.

Requirements

• Python 3.6+
• Jupyter Notebook (https://jupyter.org/install)

Running

Run script.ipynb in Jupyter Notebook

Data

The data used to train the algorithm was provided by Backblaze company (https://www.backblaze.com/b2/hard-drive-test-data.html)

Documentation

docs directory (only in Polish)

Example (prediction result for 14 days)