Reliable and continuous measurement of sediment load in rivers is essential for understanding and managing sediment transport dynamics. However, traditional sediment load monitoring methods depend on laborintensive field sampling techniques, leading to limitations in spatiotemporal coverage. This study introduces a novel approach for machine learning-aided real-time suspended sediment concentration monitoring using a horizontal acoustic Doppler current profiler. Support vector regression models for suspended sediment concentration estimation are derived from eight flow monitoring stations with sediment sampling data. A new model selection framework is proposed, integrating a global optimization algorithm in order to calibrate input variables and hyperparameters simultaneously. The three-fold cross-validation score is utilized to enhance the generalization accuracy and reliability of sediment load estimation. This work assessed the impact of considering additional input variables in addition to the backscattering signal. The findings confirm that incorporating flowrate, water level, and their time derivatives during model training improves predictability. Spatial differences between flow monitoring and sediment sampling locations significantly influence the model accuracy. This study also discusses the potential application of the H-ADCP-based monitoring system to obtain suspended and total sediment loads at an arbitrary flow monitoring station. By offering real-time monitoring capabilities and reduced reliance on labor-intensive field sampling, this work contributes to the advancement of sediment transport data acquisition methods.