Handle Missing Sensor Data

Scenario

You are building a model from aircraft health and maintenance records to predict whether an aircraft will require an unscheduled maintenance action soon. The training data includes operational, inspection, and component-level features, but many fields are missing because some sensors are unavailable on certain aircraft and some inspections are not performed on every cycle.

Question

How would you approach handling the missing data while training a reliable model for this prediction task? Explain how your strategy would depend on why values are missing and how you would validate that the approach improves performance without causing leakage.