<p>This study presents an innovative machine learning-based framework for optimizing the design of explosive charge arrangements in soil explosive compaction. The focus of this research is on using the Random Forest Regression model to predict soil compaction caused by controlled explosions, utilizing various geotechnical and environmental parameters such as grain size, moisture content, porosity, temperature, burial depth, and explosion layout geometry. A total of 30 models were developed, each representing different explosive layouts and parameter combinations, with the goal of achieving accurate predictions of soil compaction. This study highlights the capability of the Random Forest model in simulating complex and nonlinear relationships between input parameters and the resulting compaction. The results showed that burial depth and layout geometry have the most significant impact on compaction, with denser and deeper layouts, such as those in Model 11, producing the highest compaction. In contrast, layouts with less dense and more scattered arrangements, such as Model 18, resulted in lower compaction. The study also demonstrates the advantage of machine learning techniques over traditional empirical methods, offering more accurate predictions with better generalizability. Abaqus software was used to simulate the models, and results of compressive stress and settlement due to explosion loading were obtained. These results include the analysis of changes in compressive stress, settlement, and resulting compaction under the influence of controlled explosions, providing a more accurate evaluation of the effect of explosive layout on soil compaction. This model serves as a practical tool for engineers, enabling them to optimize explosive compaction strategies at the early stages of geotechnical project design.</p>