A Real-Time Flood Detection System Based on Machine Learning Algorithms with Emphasis on Deep Learning

A flood is expressed as water overflowing onto the
ground, that usually is dry, or an increase of water that has a
significant impact on human life, and it is also declared as
one of the most usual natural phenomena, causing severe
financial damage to goods and properties, as well as
affecting human lives. However, preventing such floods
would be useful to the inhabitants in order to get sufficient
time to evacuate in the areas that might be susceptible to
floods before they happen. Regarding the issue of floods,
numerous scholars proposed different solutions, for instance,
developing prediction models and building a proper
infrastructure. Nevertheless, from an economical
perspective, these proposed solutions are inefficient for
people in countries like Somalia, for instance. Hence, the
main objective of the present research paper is to propose a
novel and robust model, which is a real-time flood detection
system based on Machine-Learning-algorithms and Deep
Learning; Random Forest, Naive Bayes J48, and
Convolutional Neural Networks that can detect water level
and measure floods with possible humanitarian
consequences before they occur. The experimental results of
this proposed method will be the solution to forth mentioned
problems and conduct research on how it can be easily
simulating a novel way that detects water levels using a
hybrid model based on Arduino with GSM modems. Based on
the analysis, the Random-Forest algorithm outperformed
other machine learning models regarding the accuracy
compared to the alternative classification methods with
98.7% of accuracy. In contrast, 88.4% and 84.2% were
achieved using Naive Bayes and J48, respectively. On the
other hand, using a Deep Learning approach achieved 87%
of accuracy, showing overall good results on precision and
recall. The proposed method has contributed to the field of
study by introducing a new way of preventing floods in the
field of Artificial Intelligence, data mining, and Deep