The importance of accurate flood forecasting is rising as climate change makes it more difficult to predict when and where floods will occur. Flood predictions use supervised machine learning models, and flood levels are being tracked via the Internet of Things (IoT). Despite advancements in data-driven hydrological forecasting, producing reliable flood forecasts remains challenging. This research paper explores the possibility of utilizing adaptive deep learning with weak supervision to predict flooding in coastal cities that are equipped with Internet of Things technology. This approach uses weakly supervised learning for data labeling and enhancement, a bidirectional long-short-term memory (LSTM) RNN for time-series prediction and classification, and adaptive learning to handle seasonality. The results of the proposed system show that flood forecasting accuracy could be improved by 40% compared to the baseline models.