In the standard approach of machine learning (ML)-based river water level modelling, the lag time-based input is best served as the model input. Such a classical approach is often unable to provide significant prediction accuracy due to the presence of noise in the data. In order to address the aforementioned limitation, a hybrid particle swarm optimization-support vector machine (PSO-SVM) framework incorporating advanced preprocessing techniques is developed. Two widely used signal processing techniques, including Hodrick-Prescott (HP) filtering and empirical mode decomposition (EMD), are adopted to minimize the noise of the data and prepare the model input feature. PSO-SVM is used to develop the river water prediction model and explore its efficacy in hybrid with the advanced pre-processing techniques for enhanced prediction of river water level. The Pussur River water level data for 1 year at an interval of 1 h at the Hiron Point station in Bangladesh is used for the model development and demonstration in the current study. Each model is assessed through several statistical model performance evaluation criteria. The results demonstrate that the HPF-SVM and EMD-SVM produce good predictions of river water level with the coefficient of determination (R2) values of 0.9 and 0.89, respectively. The results conclude that the proposed framework has high potential for improved prediction of short-term river water level, particularly where the data is noisy.