A Data Scientist is building a predictive maintenance model for a fleet of vehicles. They have two tables in their feature store:
1. A sensor_readings feature table with IoT data (e.g., engine_temp,
oil_pressure) streamed continuously. This is a time-series table with
vehicle_id as a primary key and ts as a timestamp key.
2. A maintenance_logs ground truth table that records when a vehicle
component failed. This table includes vehicle_id and the exact
failure_ts timestamp.
The goal is to create a training dataset by joining sensor_readings to maintenance_logs to train a model that predicts failures. They want to join the feature data with the ground truth data to ensure point-in-time correctness and prevent data leakage during model training.
Which approach will do this?