Floods are among the most catastrophic natural disasters, causing widespread devastation to infrastructure, economies, and human lives. Floods can be triggered by heavy rainfall, rapid snow melt or dam failures. In recent years, the frequency and severity of floods have been exacerbated due to rapid climate change. As the severity of floods increases, efficient flood alert and management systems become more and more critical than ever. This chapter proposes a learning based design using data from select stations of the Mahanadi River Basin rain gauge network, aimed at predicting floods in the Mahanadi plains. We analysed 42 years’ worth of data from select rain gauge stations: Kantamal, Kesinga, Salebhata and Tikarpara. Tikarpara lies downstream of the Mahanadi, while the other stations are further upstream. Using the water level, gauge and discharge levels at upstream stations, we can predict the discharge level at Tikarpara thereby effectively predicting a flooding event downstream. Furthermore, using the decision tree, we found that the gauge and water level of Tikarpara and gauge of Kantamal, lying at the effective rain gauge network’s centre to predict water discharge at Tikarpara, was sufficient to determine flooding without depending on the attributes of the other two stations. So, in the data’s absence from the other 2 stations due to unforeseen circumstances, Tikarpara and Kantamal data is sufficient for prediction. We use 42 years’ worth of data using a random split of 70/30 for train/test RMSE and MAE for performance. Following a comprehensive examination of the patterns derived from the hydrodynamic dataset of the Mahanadi river basin, it has been determined that traditional, advanced, and AI-integrated hybrid models, including ARIMA, ARNN, ARIMA-Theta, and ARIMA-ARNN, are insufficient in fully capturing the characteristics of non-Gaussian behavior, non-linearity, and non-stationarity. To address this limitation, a BiLSTM-based model has been proposed to accurately forecast future discharge levels and enhance streamflow prediction capabilities. The performance of the key RG network at Tikarpara has been evaluated using various methodologies, including decision trees, single-layer ANN, two-layer hidden ANN, regression analysis, DNN, and Quantum-based BiLSTM models. The lower values of RMSE and MAE indicate that our approach presents a viable solution for water management and streamflow forecasting. Notably, the Quantum-based BiLSTM model demonstrates superior performance in predicting water discharge at the Tikarpara gauge station within the Mahanadi river basin.