Artificial Intelligence and GIS in Civil Engineering: A Case Study for Landslide Susceptibility Mapping in the Visakhapatnam Urban Area

Abstract

The accelerated urban expansion of Visakhapatnam, a key maritime industrial center in India, has generated intricate geotechnical hurdles, specifically regarding the structural integrity of its mountainous fringes. As urban footprints extend, built environments are increasingly penetrating the Eastern Ghats—a geological corridor defined by highly weathered Khondalite formations and precipitous elevations. These landscapes possess an innate vulnerability to mass wasting, a hazard profoundly magnified by the local tropical savanna climate and its associated extreme monsoonal precipitation. Moreover, human-induced stressors including widespread building activity, clearing of vegetation, and the disruption of hydrological runoff have undermined the natural equilibrium of these slopes. Conventional geotechnical evaluations, though technically sound, are frequently constrained by their localized nature and high resource requirements, rendering them inadequate for modeling the complex, non-linear synergies of environmental drivers across extensive municipal territories. Consequently, a critical requirement exists for a broad-scale, computational methodology to anticipate landslide hazards and safeguard the stability of the city’s expanding infrastructure.

This research bridges this analytical void by proposing an advanced, multi-disciplinary framework that synthesizes the geospatial processing capabilities of Geographic Information Systems (GIS) with the predictive precision of Artificial Intelligence (AI). To construct a dependable Landslide Susceptibility Map (LSM), the Random Forest (RF) algorithm was implemented—an ensemble learning technique preferred for its capacity to interpret high-dimensional variables while mitigating the risk of model overfitting. The predictive engine was calibrated using an extensive spatial inventory of prior landslide events alongside twelve geo-environmental parameters, such as topographic gradient, land use patterns, rock composition, and infrastructure proximity. Empirical validation demonstrated a high level of model fidelity, achieving an 88% accuracy rate and an Area Under the Curve (AUC) value of 0.91. The final LSM classifies the region into five hierarchical risk tiers, revealing that nearly 15% of the territory—clustered mainly within the Kailasagiri and Simhachalam ranges—is categorized as high-risk. This cartographic output functions as a decisive governance tool, empowering urban strategists and engineers to enforce rigorous zoning laws, optimize the deployment of reinforcement structures, and establish resilient urban growth frameworks.