Analysis of Bayesian Regularization and Levenberg–Marquardt Training Algorithms of the Feedforward Neural Network Model for the Flow Prediction in an Alluvial Himalayan River

Abstract

The purpose of this study was to develop new flood prediction models using Machine Learning algorithms for an alluvial river Jhelum in the Himalayan basin of Kashmir region, India. A major flood in the Kashmir Valley occurred in September 2014 due to the multifarious interaction among atmospheric disturbances. The study targeted the development of two Artificial Neural Network models—Bayesian Regularization and the Levenberg–Marquardt Neural Networks for flood prognosis in the Jhelum River. The models were evaluated and curves graphed of actual versus predicted discharges for 20% of the data. Levenberg–Marquardt Neural Network model performed better by predicting the flood discharge with the highest accuracy when tested for 2014 floods. With a mean squared error of 0.002128 and coefficient of determination of 95.839%, Levenberg–Marquardt Neural Network model proved to be the superior model than the Bayesian Regularization model. These models can be used for anticipating the flooding and the Department of Irrigation and Flood Control of the State of Jammu and Kashmir can take advance precautionary measures rather than issuing warnings and taking anticipatory actions based on shooting up of the flood hydrograph (norm as of now).