COMPARATIVE ANALYSIS OF GRADIENT BOOSTING MACHINES AND LONG SHORT-TERM MEMORY NETWORKS FOR STREAM FLOW FORECASTING

Shaikh, Ayub; More, Satish; Bhirud, Yuvaraj; Shelar, Vaibhav; Bagwan, Khalid

LARHYSS Journal
Volume 21, Numéro 2, Pages 7-32
2025-04-12

Comparative Analysis Of Gradient Boosting Machines And Long Short-term Memory Networks For Stream Flow Forecasting

Authors : Shaikh Ayub . More Satish . Bhirud Yuvaraj . Shelar Vaibhav . Bagwan Khalid .

Abstract

Stream flow forecasting is essential for effective water resource management and flood prediction, but it poses challenges due to the complex nature of hydrological systems. Traditional methods struggle to capture temporal dependencies and nonlinear relationships within the data, leading to inaccuracies in predictions. This study compares the accuracy of long short-term memory (LSTM) networks and gradient boosting machines (GBMs) in predicting stream flow in the Garudeshwar watershed of the Narmada River basin in central India. GBMs employ ensemble learning techniques, while LSTMs utilize recurrent neural network architectures to capture temporal dependencies. Across several datasets (training, validation, and testing), GBMs consistently beat LSTMs through thorough examination utilizing measures including MAE, RMSE, R², and RMSPE. GBMs demonstrate superior accuracy, precision, and reliability in modeling stream flow dynamics, showcasing their effectiveness in water resource management and flood prediction endeavors. The goal of the project is to improve methods for forecasting stream flow and support well-informed decision-making in efforts to manage water resources and prevent flooding.