Ensemble of Multi-Regression Learning Methods for Iot-Enabled Early Flood Prediction

Abstract

Flooding is an utmost devastating natural fact that has a terrific effect on earthlings, living
organisms and infrastructure. Prediction of flood in an early stage creates a better
circumstance in and around the abode. In this regard, quite a number of research works have
been investigated and implemented using recent techniques. However, this field still paves
way for further research in terms of precision and dependability due to the dynamicity of
weather conditions and rainfall intensity. Internet of Things (IoT) and Machine Learning
(ML) models are in their period of development for flood prediction. This paper utilizes
historical and IoT sensor weather data with ML regression algorithms (LASSO Regression
(LAR), Ridge Regression (RR), Linear Regression (LR), Support Vector Regression (SVR)
and Random Forest Regression (RFR)) to anticipate flood occurrence based on the severity of
rainfall. The main objective is to enhance the accuracy of rainfall prediction related to early
flood detection using ensemble multi regression learning method (EMRLM). Two different
sets of data are engaged uniquely to predict rainfall in two different ways using machine
learning algorithms. Error-rate-based weighted prediction results of best performing
algorithms are merged. Final outcomes derived from both approaches are fused to further
improve prediction accuracy. Depending on the intensity of rainfall, flood occurrence is
classified as possibility of flood and no possibility of flood. Regression evaluation metrics
handled proves the effectiveness of the recommended model. Results obtained with RMSE
(0.16966), MSE (0.028785), MSLE (2.42E-06), MPD (0.00026631) and MGD (2.46E-06)
portrays that the combined work excels the effort of individual learning methods in providing
a reliable flood prediction model.