The Potential of Machine Learning for Tackling Flood Disaster in Monrovia

Flooding is, and will likely continue to be, the most catastrophic natural event that undermines successful urban development in the developing world. Because of their low elevation and proximity to frequent windstorms, coastal communities are particularly more susceptible to natural disasters. The current study uses 40 years of terraclimate data to examine the possibilities of employing machine learning algorithms to predict flood incidences in the capital of Liberia, Monrovia. Two models of Artificial Neuron Networks were selected: Elman Neuron Network, and Cascade Feedforward Neuron Network. The models were trained and tested using two-layer network architecture, different numbers of neurons, and alternating transfer functions (TANSIG and LOGSIG). Five flood-inducing variables: precipitation, windspeed, minimum temperature, maximum temperature, and soil moisture, as input variables. Runoff was chosen as the output variable. The Coefficient of Determination (R2) and Root Mean Square Error (RMSE) were used to test the accuracy of the models. From the results, it was found that the Cascade Feedforward Neuron Network was the most stable during the testing phase, as measured by the statistical coefficient of determination. The CAS-1 model with its ten neurons, two hidden layers, and TANSIG transfer function yielded 51.5% when all five inputs were fed into the network, followed by the Elman-3 model with an R2 score of 37.8%. For the testing phase, the Elman-2 model had a Root Mean Square Error of 0.061516, while the CAS-3 model had a value of 0.068528.