Using a Deep Learning Model to Predict Flash Flood Likelihood

Overview

A flash flood is when an area floods due to sudden extreme rainfall or another natural condition. Because of their destructive power, many families lose their homes yearly due to flash flooding. Creating an AI model to detect flash floods would assist these people, as they would have time to evacuate and protect themselves. The engineering goal of this project was to design, build, and test a deep-learning model to predict the likelihood of a flash flood given monthly rainfall data for the past year to determine whether a flash flood will occur in the next month given rainfall trends that residents in flood-prone areas can use to evacuate safely and with enough time with an accuracy of at least 95%. The AI model created consisted of an artificial neural network (ANN) and a logistic regression (LR) model. The model was created in a Python Jupyter Notebook in Google Colab, written in Python version 3.10. The dataset was taken from Kaggle, uploaded by the user Mukul. The independent variable was the monthly rainfall amounts for the past year, and the dependent variable was the likelihood of a flash flood in the next month. The models were created, trained, and then tested using the raw data. This creation and training involved having an ANN with eleven layers, alternating between dense and dropouts. These layers increased the complexity of the model. The ANN was checked over by the LR, which if certain accuracy minimums were not met, recreated the ANN. The ANN’s final accuracy was 97.51% and the LR model’s final accuracy was 98.33%. Thus, the model exceeded the engineering goal of at least 95% accuracy, satisfying the design criteria.

Using a Deep Learning Model to Predict Flash Flood Likelihood

Overview

Competition history

Related projects

The Effect of Atmospheric Conditions on Flash Flood Prediction Using Deep Learning

Development of a Novel AI Last-Minute Flood Forecasting and Warning Gadget Using IOT and Deep Learning

Mitigating Flood Risk by Predicting River Gauge Heights Using a Neural Network

FloodCast: Predicting Floods and Their Socioeconomic Impacts Through Deep Learning Neural Networks and Regression

FloodCast: Real-Time Flood Mapping and Prediction in Southeast Asia Using Remote Sensing Data

Real-Time Probabilistic Delineation of Flood Warning Polygons to Reduce Flood Alert Fatigue

Augura: Flood Risk Prediction Using Machine-Learning and Geographic Information Systems

Low-Cost and Easily Accessible Flash Flood Detection System

Browse more like this