Practicing on the PB practice data set for the upcoming challenge
- Pitney Bowes is one of the leaders in the Mailing meters space.
- Mailing meters allow businesses to be more productive by simplifying their mailing and shipping.
- We offer leased hardware, as-well as SaaS solutions.
- Our clients must be able to rely on our meters to function to meet their customers’ needs.
- Pitney Bowes uses predictive maintenance to reduce the risk of down-times of meters deployed at our clients’ offices.
- Since our meters are part of our clients’ business operations, we must be proactive about identifying potential problems early on to avoid any sort of disruption for our clients.
- Predictive maintenance allows us to have the required replacement parts where they are needed, and to proactively schedule an appointment with our customers if needed.
- The meters are Cloud-connected, which allows us to monitor the health of a device and detect potential problems early on.
- For this challenge, we generated a training sample for ~40k meters. You are asked to build a model that can predict which meters will fail within the next 7 days.
- A csv file with the meter id and the fail_7 column forecast for the meters in the test set.
- Python notebook with code and sufficient documentation.
- Single page poster summarizing work done and techniques implemented which will be used for presentation on the final day.
- Video presentation for the work done.
Problem: Consumers complain about device failures that had been installed months ago.
- Reason 1:
- Unable to idenitfy device failures in advance.
- Reason 2:
- Not enough data to make a prediction to detect meters that may fail tomorrow.
- Reason 3:
- Data is available but engineering team designated to fix is not communicated.
- Reason 4:
- Data and engineering team is aware, but consumer hasn't responded or been pro-active about their meters.
- Reason 5:
- Lack of awareness that the device installed in the past needs maintenance.
- Root cause
Team 2