Replication Code and Data for "Explainable Machine Learning Reveals Shifting Drivers of Mortuary Differentiation: A 1,500-Year Sequence from Liuwan, Northwest China"

This repository contains the replication code and data for the analysis presented in the accompanying article. The dataset comprises 1,485 burials from the Liuwan Cemetery (柳湾墓地), Ledu District, Qinghai Province, China, spanning three cultural phases of the late Neolithic through early Bronze Age (c. 3300–1500 BCE): Banshan (半山), Machang (马厂), and Qijia (齐家). We apply gradient-boosted classification (XGBoost) with SHapley Additive exPlanations (SHAP) to identify the material variables that most strongly differentiate mortuary treatment across and within cultural phases, and to track how these drivers shift over a 1,500-year sequence. The analysis is complemented by PCA, Bayesian Gaussian Mixture Modeling, UMAP visualization, and PELT change-point detection.