Author(s)

Mingfei Zhou

Corresponding Author

Mingfei Zhou

Abstract

This paper develops a quantitative framework to predict the spatial–temporal diffusion of the printing press across cities in the Holy Roman Empire. Building on a newly curated dataset of city‑level covariates (urbanization as a proxy for literacy, population, GDP per capita, Hanseatic membership, confessional affiliation) and the year of first press installation, we formalize adoption timing as a supervised regression target—the delay in years relative to Mainz (1452). After exploratory analysis confirms weak wealth–delay associations but a strong geographic gradient, we propose a CatBoost‑based pipeline that natively handles mixed data types and non‑linear interactions. Baseline models (linear regression and k‑nearest neighbors) under a temporal holdout (train ≤1500, test >1500) capture limited structure, motivating the shift to gradient boosting and feature interactions. Using a synthetic but historically consistent expansion with explicit distance‑to‑Mainz and language categories, CatBoost attains substantially lower errors (≈3.3‑year MAE) and high explanatory power (R²≈0.81). Global importance shows that spatial frictions (log distance) dominate, while urbanization and population provide strong demand‑side signals; religion and trade‑network status add meaningful context. The approach yields city‑level narratives (e.g., Vienna vs. Cologne) that connect predicted delays to interpretable factors. The framework is readily transferable to real data once distances, university proximity, and terrain/river barriers are integrated, and it generalizes to other historical and modern diffusion problems.

Keywords

Printing Press Diffusion, Holy Roman Empire, Technological Adoption, Catboost