Local scour around buried pipelines in riverbeds is one of the major factors contributing to bed instability and the vulnerability of fluid‑transport infrastructure. Incorrect estimation of scour geometry can lead to significant technical, economic, and environmental damages. Therefore, the development of reliable methods for predicting scour hole geometry plays an important role in the safe and economical design of such structures. Recent studies have shown that machine learning techniques and artificial intelligence algorithms have attracted considerable attention as effective tools for analyzing and predicting complex hydraulic phenomena. In this study, the performance of three machine learning approaches, namely Artificial Neural Networks (ANN), Support Vector Machines (SVM), and the k‑Nearest Neighbors (KNN) algorithm, was evaluated for predicting the geometry of local scour holes, including maximum scour depth and maximum scour length, around buried pipelines in the presence of an impermeable plate. The analysis was conducted using 64 sets of laboratory experimental data. The hydraulic and geometric input parameters included the flow Froude number, the ratio of initial burial depth of the pipeline to channel width (e/B), and the ratio of pipeline diameter to channel width (D/B). The results indicated that the ANN model achieved more accurate and stable predictions than the other methods, with coefficients of determination greater than 0.96 for predicting both scour depth and scour length, while the SVM and KNN models showed comparatively lower accuracy. Sensitivity analysis also revealed that the ratio of initial burial depth to channel width (e/B) is the most influential parameter affecting local scour geometry. Overall, the findings demonstrate that machine learning approaches, particularly artificial neural networks, can serve as effective tools for predicting scour and improving the design and safety of buried pipelines in riverbeds.