Trends and innovations in hybrid backpropagation–trust region quasi-newton neural network algorithms for intelligent flood detection systems

Abstract

This study conducts a Systematic Literature Review (SLR) to examine recent trends and innovations in the application of the Backpropagation–Trust Region Quasi-Newton (BP–TRQN) algorithm in intelligent flood forecasting systems. The literature was systematically collected from Scopus, Dimensions, and Google Scholar, with the search restricted to publications from the last ten years. The selected studies were analyzed through a rigorous process of article screening, data extraction, and bibliometric as well as content analysis using VOSviewer and RStudio. The review indicates that the integration of BP–TRQN within hybrid neural architectures, particularly ConvLSTM and CNN–RNN ensembles, is associated with approximately a 5–15% improvement in forecasting accuracy, as reflected in reductions in RMSE and MAE, compared with conventional first-order optimization methods such as GD, SGD, and Adam. In addition, the number of iterations required to achieve convergence was reduced by approximately 20–40%, suggesting improved computational efficiency and greater model stability under non-stationary and noisy hydrometeorological conditions. These findings suggest that BP–TRQN has considerable potential as a robust optimization framework for improving the reliability, efficiency, and adaptability of flood forecasting systems. The main contribution of this study is the provision of a structured synthesis of temporal research trends, methodological innovations, and empirical evidence concerning the role of BP–TRQN in intelligent flood forecasting, thereby identifying current research gaps and future directions for the development of operational early-warning systems. Nevertheless, the review also identifies a significant research gap, namely the limited empirical validation of BP–TRQN within operational early-warning systems, particularly in terms of real-time forecasting performance and cost-effectiveness. Addressing this gap should be considered an important direction for future research, with potential implications for the development of sustainable and data-driven disaster management systems.