Impact of Learning rate in the development of Multi-Layer Perceptron for flash flood assessment

Abstract

Environmental and climatic changes produce adverse effects like flash floods. Flash floods are very common in many countries and they are very sudden. Sudden flash floods cause infrastructural losses and casualties. Usually high precipitation index, water color, wave’s pattern, temperature, wind velocity and environmental CO2 levels have been measured as a yardstick to determine the run-offs. Diversified methodologies have been adopted to determine the flash flood actual timings and other properties. Sensors based, radar based and satellite based observations were used to investigate the flash floods. Artificial intelligence based algorithms were developed to detect the flash floods timely with less false alarm rate. In this research paper multi-layer perceptron neural network with different learning rate has been proposed for the vigorous determination of flash floods with less false alarm rate. Variable learning rates have been adopted to achieve the higher accuracy objective function for the better results. Impact of changing the learning rate has been discussed to prove that appropriate learning rate must be used to reduce the error rate with in the links of neuron.