Training Analysis of Optimizers on Flood Object Dataset using EfficientNetB0 - A Novel Deep Learning Performance

Urban flooding is an escalating global challenge, exacerbated by rapid city growth and climate change. These inundations cause widespread economic damage, pose health hazards, and disrupt critical infrastructure. While deep learning has proven effective in various predictive tasks, its application to flood forecasting remains limited. The performance of these models hinges on their training process, optimization techniques, and the number of learning iterations. This study aims to enhance urban flood resilience. It focuses on evaluating the EfficientNetB0 model's performance using three optimization algorithms: ADAM, SDGM, and RMSPROP. The research assesses each optimizer's efficacy by varying the number of epochs while maintaining other hyperparameters constant. Notably, when applied to object detection in the training dataset, the model achieves an outstanding 100% success rate using the ADAM optimizer.