<p>Real-time flood prediction (RTFP) is essential for early warning systems. Predicting flood occurrences with longer lead times is challenging in regions with limited rain gauge coverage. The accessibility of satellite precipitation products (SPPs) provides an alternative solution to these limitations. SPPs offer global, real-time precipitation estimates and are suitable for use in RTFP models. This study compared daily SPPs from Integrated Multi-Satellite Retrievals for Global Precipitation Measurement (IMERG) and Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN) with ground-based observed rain gauge data from the India Meteorological Department (IMD) for the period 2001 to 2021 using contingency tests. Results indicate a strong correlation, with probabilities of detection at 77.81% for IMERG and 75.86% for PERSIANN. Subsequently, daily precipitation data from IMERG and PERSIANN were used to develop four machine learning-based RTFP models: Feedforward Neural Network (FFNN), Extreme Learning Machine (ELM), wavelet-integrated FFNN, and wavelet-integrated ELM. These models predict water levels (WL) at the Baltara gauging station of the Kosi River, with lead times from 1 to 10 days. The findings indicated that wavelet-integrated hybrid models surpassed FFNN and ELM. Among the standalone models, FFNN outperformed ELM providing satisfactory predictions up to a 5-day (NSE = 0.67) and 1-day (NSE = 0.65) lead time using the IMERG and PERSIANN datasets, respectively. At a 7-day lead time (NSE = 0.58) only the wavelet-integrated FFNN model performed better with the IMERG dataset, while neither hybrid nor standalone models achieved acceptable results for a 10-day lead time.</p>