Discharge Prediction at Bahadurabad Transit of Brahmaputra-Jamuna Using Machine Learning and Assessment of Flooding

Abstract

The river Brahmaputra and Jamuna have a significant contribution to water transportation, agriculture, and the livelihood of people living on the river banks. Ensuring the proper utilization of these channels, accurate water discharge prediction is a must since it can be of benefits for managing the river and allocating water resources. In this study, three different machine learning algorithms, K-nearest neighbor, decision tree, and random forest regressor, along with different hyperparameters, are presented, which have been found more efficient among the other machine learning approaches. Daily water level and maximum velocity were used as explanatory variables, and water discharge was used as a response variable. Data from 2005 to 2013 was used for training the model, and 2014 to 2019 was used for evaluation. Among these three models, the k-nearest neighbor has performed exceptionally well. This model's R2 value and mean absolute percentage error are 0.9447 and 17.38, respectively. The obtained discharge rates are further compared with previously recorded discharge data before, during, and after major floods in those regions and which are found to have a linear relationship with river flooding.