Incorporating establishment typologies into freight production models: A latent class approach beyond industry codes

This study examines how establishment typologies influence the accuracy of freight production (FP) models in two contrasting Indian regions - Kerala and Hyderabad. Two types of typologies were constructed using Latent Class Cluster Analysis: An Observable Typology, based on observable variables (commodity type, distance to city centre, proximity to freight infrastructure), and a Survey-based Typology, based on survey-derived attributes (fleet ownership, commodity value density, business age, and period of formation). FP models were developed using both typologies and benchmarked against regional and industry-class models. In Kerala, the Survey-based Typology yielded the most accurate models, with two out of five classes showing a substantial reduction in residual variance compared to regional models. These models improved R2 values by up to 22 % over unsegmented baseline models. The Observable Typology also improved model performance, but to a lesser extent. In Hyderabad, typology-based models improved prediction accuracy for specific classes, particularly those with high freight-generating intensity. Across both regions, the Survey-based Typology consistently explained freight production variation better than industry classifications or spatial factors alone. The results confirm that data-driven typologies, especially those capturing firm behaviour and logistics attributes, provide significant gains in modelling freight production and enable finer-grained understanding of establishment-level freight activity.