Question

A data scientist is building a model to predict daily energy consumption for commercial buildings. The dataset spans three years of hourly readings with clear daily and seasonal patterns. When evaluating candidate models, the scientist wants a validation strategy that produces a realistic estimate of how the model will perform on future data. What is the most appropriate way to split the data for training and validation?

Accepted Answer

Use a chronological split, training on the earlier time period and validating on the most recent, held-out time period.

Answer

Randomly shuffle all hourly records and hold out 20% as the validation set.

Answer

Apply standard k-fold cross-validation with folds assigned by random sampling of rows.

Answer

Stratify the split by hour of day so each hour is proportionally represented in both sets.