Job-housing separation (JHS) is a critical constraint on sustainable urban development, increasing energy consumption and air pollution. While planners increasingly recognize that “one-size-fits-all” interventions are insufficient, most studies fail to quantify the optimal intervention ranges derived from these non-linear relationships. This leaves planners without data-driven, actionable targets. This study bridges this gap using a replicable framework with multi-source Location-Based Services (LBS) data and interpretable machine learning (IML). It first introduces a comprehensive, bidirectional (resident and employee) JHS metric to overcome traditional biases. More importantly, it moves beyond simple driver identification by precisely quantifying the non-linear effects of built environment factors to determine their optimal intervention ranges. Using Beijing as a paradigm, results show: (1) The comprehensive, bidirectional metric (integrating resident and employee perspectives) successfully overcomes the significant biases of traditional single-perspective measures, identifying distinct JHS typologies. (2) Housing Price and Job-Housing Ratio are the core drivers of JHS, while key planning levers like bus station density and land-use mix exhibit “U-shaped” effects, indicating that “moderate”—not maximum—levels are optimal for mitigating JHS. (3) As its key quantitative contribution, the framework's diagnostic tool operationally defines context-specific strategies for different JHS typologies. For example, a typical residential zone requires increasing employment density (to 1144 jobs/km 2 ), while a typical employment center needs employment decentralization (job-housing ratio (JHR) to 0.956) and affordable housing (12,720–50,202 yuan/m 2 ). This study provides a generalizable and replicable data-driven framework for quantifying non-linear effects and deriving context-specific intervention ranges, supporting refined, low-carbon, and sustainable urban governance. • Proposes a bidirectional JHS metric overcoming single-perspective bias. • Moves beyond ML visualization to operationally define "moderate" levels. • Innovatively translates non-linear relationships into specific, recommended intervention ranges. • Develops a diagnostic framework applying quantitative strategies to JHS subdistricts.
Lin et al. (Wed,) studied this question.