A cinema company wants to build a model to predict customer visit patterns for the coming year. They have three years of customer visit data across 300 theaters; however, the data has been stored in d

Using only the last year of data discards two-thirds of the available training data, significantly reducing the model's ability to learn seasonal and long-term visit patterns and lowering predictive accuracy.

Transforming all data into a consistent format is the essential preprocessing step required before training any ML model. ML algorithms require uniform input structure to identify meaningful patterns - inconsistent formats introduce errors and bias that lead to inaccurate predictions. Standardizing three full years of data across all 300 theaters ensures the model trains on a complete, coherent dataset and produces reliable visit-pattern predictions.

Training separate models per format group is inefficient, prevents the model from learning patterns across the full dataset, and produces fragmented, inconsistent predictions that cannot be easily combined.

Standard ML models do not natively process multiple heterogeneous input formats simultaneously - data must be preprocessed into a consistent structure before training regardless of which model type is selected.