While cleansing sensor data for the predictive-maintenance model, Meridian's data engineers find that roughly 8% of readings are missing due to intermittent sensor connectivity, and the failure events in the historical log are rare compared to normal-operation readings. Which pairing correctly matches each problem to the right data-preparation response?

Data cleansing and enhancement covers exactly these two situations, and each needs its own technique: missing readings from intermittent connectivity should be handled with an appropriate missing-data strategy such as imputation or explicit flagging, preserving as much signal as possible rather than discarding records outright; the rarity of failure events relative to normal operation is a class-imbalance problem best addressed with data augmentation or balancing techniques during Data Preparation, because a model trained on mostly-normal data will struggle to recognize the rare failures it exists to predict. Deleting every record with any missing value throws away a meaningful chunk of otherwise-usable data and dodges the augmentation the rare-event problem actually needs — and imbalance is a data-preparation and training concern, not something that surfaces only at Model Evaluation. Ignoring the missing data because 8% "seems low" is an arbitrary judgment without addressing root cause, and deleting rare failure events as outliers is actively harmful — those rare events are precisely the signal the model needs to learn. Generative-AI text summarization has no bearing on numeric sensor gaps, and increasing sensor frame rate going forward does nothing to fix the historical data imbalance already in hand — both pair the wrong tool to the wrong problem.