Soil arching is an important phenomenon in geotechnical engineering because it significantly influences stress distribution and displacement patterns in earthworks. Nevertheless, the intricate behavior of soil arches, particularly in stone column–improved loose silty sand supporting fill, remains elusive. Designers encounter uncertainties in predicting stresses and settlement of stone column–improved loose silty sand. Usually, they apply cohesive soil settlement equations and Boussinesq elasticity-based stress equations, disregarding soil arching impact. This study aims to investigate the unique characteristics of this soil system by employing innovative laboratory-scale physical model tests for soil arching testing. The functional behavior of this system under varying surcharge loads and fill heights was studied. Additionally, a preliminary analytical framework was proposed. The normalized fill height relative to the stone column clear spacing is a crucial parameter. The study revealed that increasing this height enhances the soil arching efficiency, stress concentration, deformation, and column loads. All these variables increased with the applied surcharge. The study evaluated existing stress equations against the experimental results. Then, new equations for stress-settlement relationships were developed through multistatistical analysis of the experimental data using Design-Experts (DoE) software (version 12.0.3.0), tailored specifically to this soil system. Their agreement was assessed against published experimental data. The study further provides illustrative assessment charts. Together, these elements offer a preliminary framework for predicting stresses and settlements and estimating settlement improvement. This framework represents a key innovation of this study. An administrative example illustrates their use. While these insights provide a preliminary basis, their direct applicability for real-world design necessitates further comprehensive validation through full-scale testing and/or advanced numerical modeling.