Construction Suitability and Influencing Factors in Abandoned Mining Areas Based on Machine Learning: A Case Study of Luhe District, Nanjing, China

Abstract

Land disturbance and ecological degradation are widespread in abandoned mining areas, and their redevelopment is shaped by multiple factors. Identifying construction suitability patterns and key influencing factors can provide a quantitative basis for the reuse of these areas. Focusing on an abandoned mining area in Luhe District, Nanjing, this study selected 16 influencing factors. K-means clustering and a self-organizing map (SOM) were used to identify construction suitability types. Principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE) were used to visualize the cluster structure. Random forest (RF) and SHapley Additive exPlanations (SHAP) were then combined to analyze the overall contributions and interaction patterns of the individual factors. The results showed the following: (1) The study area was classified into five construction suitability types with no inherent ranking: the Soil-Location Suitability Type, Entrance Accessibility–Public Service Suitability Type, Vegetation Conservation Suitability Type, Terrain-Location Suitability Type, and Ecological Restoration Suitability Type. Among them, the Entrance Accessibility–Public Service Suitability Type and Vegetation Conservation Suitability Type were the most widely distributed. (2) The influence of the central urban area made the greatest overall contribution, followed by soil texture and vegetation coverage. Vegetation coverage had the strongest effect on the model’s ability to reproduce the cluster-derived classes under spatial validation. Vegetation diversity, entrance accessibility, soil texture, and pollution level also showed high permutation importance. (3) Factor contributions were type-specific, with soil, location, accessibility, vegetation, and pollution conditions showing distinct contribution and interaction patterns across the five types. Differentiated strategies should therefore be tailored to each type, including location and service optimization, vegetation protection and low-disturbance use, terrain management, and pollution control and ecological restoration.