Floods are prevalent and recurring catastrophic natural disasters that disrupt economic and social services. Different measures like inundation modelling, flood forecasting, and flood hazard and risk mapping can be adopted to attenuate the adverse effects and economic losses of floods, thus identifying flood-susceptible zones. One of the most commonly used flood forecasting systems is the physically based model where dynamicity and physical characteristics of the watershed will be considered. These models are encountered by complexities in hydrological processes and non-linear characteristics of the input parameters. The reliability of these models is affected by the purpose of modelling, the nature of the watershed, and the quality of input variables like rainfall, land use land cover, and spatiotemporal variability of inputs. They also suffer from computational instabilities and long runtime problems, which are particularly important in real-time applications. Machine learning models are becoming a remedy to counteract this issue as the speed of computation of these models is more than the other. In this study, a comparison was done between coupled models based on the HEC-HMS rainfall-run-off and the HEC-RAS hydraulic routing model with two machine learning-based flood models - Hybrid wavelet Artificial Neural Network model (WANN) and Hybrid Wavelet Support Vector Machine model (WSVM) for different lead times in the Achankovil river in Kerala. The hourly water level at the Konni gauging site was simulated with one-hour lead times. The performance of each model structure was evaluated based on standard performance criteria and it was evident that WSVM performed better than WANN and HEC-RAS simulations.