This study applied machine learning to Lagrangian particle dynamics to provide a detailed understanding of the effects of large-scale coastal development on material transport patterns. The approach contributes to improved environmental assessment and sustainable marine planning. The study area, Saemangeum, a semi-enclosed estuarine system undergoing simultaneous land reclamation and port construction, was modeled under three development scenarios: pre-, mid-, and post-development. Five trajectory-based features (total travel distance, displacement ratio, duration, radial average, and curvature) were defined to quantify the structural complexity of particle motions. These features were analyzed using principal component analysis and k-means clustering to classify the flow pathway types. The results showed that, although a mixture of movement patterns appeared in the pre-development scenario, the proportion of recirculating-type particles increased over time, reaching 68.1% in the post-development scenario. In contrast, the proportion of dispersive type particles markedly decreased. Despite these shifts in type frequency, the mean feature values for each type (e.g., travel distance < 98,000 m for advective particles, residence time ~28 days for recirculating particles, and radial average > 6,500 m for dispersive particles) remained consistent across scenarios, indicating the robustness and generalizability of the proposed classification scheme. By focusing on trajectory-based dynamic patterns rather than static metrics, such as velocity fields or concentration distributions, this study offers a novel framework for diagnosing structural changes in transport processes due to coastal development. Integrated machine learning-based classification and spatial distribution analysis provide practical insights for future coastal environmental impact assessments and water quality management strategies.