Replication package: Multi-slice training for bank customer churn prediction

Replication package for the article "Does Multi-Slice Training Improve Bank Customer Churn Prediction? Evidence from Corporate and Retail Transaction Histories" (Kodirov, Meliboev, Muminova, Khasanov and Temirov; submitted to FinTech, MDPI, 2026). The article tests whether training churn models on several monthly snapshots of each customer (multi-slice training) improves prediction over training on the latest snapshot only, using corporate clients of Sber (MBD-mini) and retail clients of T-Bank (TTRS), with the Berka dataset of a Czech bank as a supplementary case of partial defection. It also compares out-of-time, customer-disjoint and random row-wise validation, and documents the growth of remote banking in Uzbekistan from Central Bank of Uzbekistan statistics and the Global Findex Database. The package contains the analysis plan written before any model was trained on the bank data (Uzbek original with its SHA-256 hash and an English translation) and the log of deviations from it; the Python code that produces every number, table and figure in the article, with package versions; the derived monthly customer-slice tables with features and churn labels for MBD-mini, TTRS and Berka; a panel of registered users of remote banking systems by bank in Uzbekistan, January 2018 to October 2025, compiled from the monthly tables published by the Central Bank of Uzbekistan; the country-level Global Findex 2025 file; all result files, including test-set predictions, AUC, PR-AUC and EMPC values, DeLong comparisons, robustness checks and SHAP importances; and the figures of the article. A SHA-256 checksum is provided for every file. Raw transaction files are not included; they are available from their original sources listed in the README. The Global Findex microdata for Uzbekistan are not included because the World Bank Microdata Library terms do not permit redistribution. Data sources: MBD-mini (Sber AI Lab, CC BY 4.0), TTRS (T-Bank, CC BY-NC-SA 4.0), Berka financial dataset (CTU Prague Relational Learning Repository), Central Bank of Uzbekistan payment system statistics, Global Findex Database 2025 (World Bank, CC BY 4.0). Because the package contains material derived from TTRS, it is released under CC BY-NC-SA 4.0.