Objectives
Indicate reboot probability using an XGBoost algorithm.- Identify detailed reasons for reboots and suggest preventative steps.
- Cluster devices to train specialized models for more accurate results.
- Created an ML model achieving >80% precision and >85% accuracy.
- Predicted reboots 12 hours in advance, giving operators time to take preventive action.
- The model was deployed in production, with big data tools used to handle 10M+ devices per hour.
- Real-time predictions and insights were made available via Tableau dashboards.
- Cloud infrastructure was critical in handling the large-scale data involved:
- One day of raw data: ~100GB
- Total dataset size: 16 TB
- Number of features: >500 (Data was collected from each device every few minutes)