Scholarlink

<p>Interpretations of Palaeolithic landscapes have frequently relied upon a priori site typologies, functional classifications, and environmentally deterministic frameworks that risk circular reasoning, typological reification, and underdetermined behavioural inference. This study develops and applies an explicitly inductive framework for identifying behavioural landscape structure without imposing pre-defined functional or categorical assumptions. Using a multivariate dataset of 158 georeferenced open-air Palaeolithic occurrences from the Vidarbha region of eastern Maharashtra, India, the analysis integrates topographic, hydrological, and raw-material accessibility variables alongside a kernel-derived archaeological persistence measure within an unsupervised spatial learning workflow. Unsupervised spatial learning refers here to the application of clustering algorithms requiring no predefined class labels to spatially referenced archaeological datasets, distinguishing this exploratory inferential mode from confirmatory supervised prediction. Cluster robustness was evaluated through multiple internal validation diagnostics, including elbow analysis, silhouette optimisation, gap statistic assessment, sensitivity testing, and cross-algorithm comparison. Results reveal reproducible multivariate landscape configurations best understood as probabilistic behavioural landscape systems rather than discrete functional site types or deterministic settlement categories. These findings demonstrate that coherent archaeological landscape organisation can be detected inductively prior to typological assignment, while clarifying persistent zones of temporal palimpsest formation, proxy limitation, and interpretive ambiguity inherent to open-air records. The study argues that unsupervised spatial learning is most productive when positioned as an epistemic support framework that formalises pattern discovery, stabilises inductive reasoning, and preserves the interpretive centrality of archaeological expertise, contributing to a more transparent and methodologically accountable approach to behavioural inference in Palaeolithic landscape research.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Inferring Behavioural Landscape Systems in Palaeolithic Archaeology: Unsupervised Spatial Learning as a Framework for Formal Inductive Inference

  • 期刊论文

  • 发表日期：

    2026年7月17日

查看全文
  • Sushant Begade

摘要

Interpretations of Palaeolithic landscapes have frequently relied upon a priori site typologies, functional classifications, and environmentally deterministic frameworks that risk circular reasoning, typological reification, and underdetermined behavioural inference. This study develops and applies an explicitly inductive framework for identifying behavioural landscape structure without imposing pre-defined functional or categorical assumptions. Using a multivariate dataset of 158 georeferenced open-air Palaeolithic occurrences from the Vidarbha region of eastern Maharashtra, India, the analysis integrates topographic, hydrological, and raw-material accessibility variables alongside a kernel-derived archaeological persistence measure within an unsupervised spatial learning workflow. Unsupervised spatial learning refers here to the application of clustering algorithms requiring no predefined class labels to spatially referenced archaeological datasets, distinguishing this exploratory inferential mode from confirmatory supervised prediction. Cluster robustness was evaluated through multiple internal validation diagnostics, including elbow analysis, silhouette optimisation, gap statistic assessment, sensitivity testing, and cross-algorithm comparison. Results reveal reproducible multivariate landscape configurations best understood as probabilistic behavioural landscape systems rather than discrete functional site types or deterministic settlement categories. These findings demonstrate that coherent archaeological landscape organisation can be detected inductively prior to typological assignment, while clarifying persistent zones of temporal palimpsest formation, proxy limitation, and interpretive ambiguity inherent to open-air records. The study argues that unsupervised spatial learning is most productive when positioned as an epistemic support framework that formalises pattern discovery, stabilises inductive reasoning, and preserves the interpretive centrality of archaeological expertise, contributing to a more transparent and methodologically accountable approach to behavioural inference in Palaeolithic landscape research.