ts / Timur Salakhetdinov

P-002 · 2025 · Team project

Hierarchical Bayesian Pareto/NBD customer modelling

A team replication of Abe (2009), implementing Bayesian customer-activity forecasts on the CDNOW transaction dataset.

PythonBayesian modellingMCMCCDNOW

Contribution and context

I worked on implementing and validating the hierarchical Bayesian Pareto/NBD model from Abe (2009). This team project uses the public CDNOW transaction dataset to estimate customer activity and future purchases.

Method

The workflow prepares customer-level transaction summaries, estimates models with and without a sales covariate using MCMC, and compares calibration and validation performance at customer and aggregate levels.

Public outputs

The repository documents descriptive statistics, model-fit tables, parameter estimates with credible intervals and customer-level posterior summaries. Outputs include expected lifetime, survival probability and forecast transactions, alongside figures for model behaviour and predictive fit.

Scope

This is a methodological replication on a benchmark dataset. The public outputs demonstrate the implementation and evaluation workflow; they do not establish performance for a production customer base.

View public code and documentation on GitHub ↗

Timur Salakhetdinov ·