BNDNN: Batch Normalization Based Deep Neural Network for Predicting Flood in Urban Areas

8 Disaster is a very serious dissipation that arises for a short time period, but the impact of 9 that disaster on human society is very dangerous and very long-lasting. Disasters are 10 categorized into two types like natural disasters and manmade disasters. Among all 11 disasters, of all the natural disasters, flood is the commonplace natural disaster. Flood 12 disaster that causes huge loss of human life, diversity as well as economic loss, which is 13 very dangerous for the developing countries and developed countries also. Nowadays 14 during the monsoon season flood is dangerous for all the geographical areas located 15 nearby water bodies. Much research has been done for flood detection. Machine Learning 16 and many other recent technologies are playing a vital role in predicting the occurrence of 17 floods. For prediction purposes, a huge amount of data is requiring collected from sensors 18 deployed in various locations. In this paper, we used the Batch normalization with Deep 19 Neural Network (BNDNN) technique for the classification of data in three classes as 20 Low, Moderate, and High. The result obtained from our proposed model is compared 21 with some other models like Decision Tree (DT), Support Vector Machine (SVM), 22 Artificial Neural Network (ANN), and Deep Neural Network (DNN). In this our 23 proposed BNDNN provides 89% accuracy which is higher among all existing models. 24 Models are compared based on some parameters like Accuracy, Precision, Recall, F – 25 Score. The compression among all the models used in this paper shows that our proposed 26 model provides better results.

the catchment areas.  In the current section of the paper, we provide the basic introduction of the flood disaster 67 as well as some statistical data regarding the flood. These sections also highlight the basic 68 information of the model which we use in our paper. Paper flow is in section 2, we 69 provide a literature survey. In this section, we discuss the previously used model and 70 algorithm used in flood prediction. In section 3, We have discussed some of the standard 71 algorithms and models used earlier in flood prediction. In section 4, we discuss our 72 proposed model for flood prediction. In section4, we provide information regarding the 73 data set we used in our paper as well as compare our model with some other models. In 74 section 5, we provide the conclusion of the paper as well we the future scope. 75 Our major contribution in this paper: 76  In this paper, we introduce the new approach, where the batch normalization 77 approach combines with DNN for meat flood event prediction. 78  The accuracy of our model is also compared to some existing standard models 79 and algorithms and our proposed model provides better prediction accuracy. 80  In this paper, we consider datasets with different environmental parameters, such 81 as cloud cover, precipitation, average. temp, min. temp, max. temp. along with 82 this parameter year, the month is also included. The study area in this paper is disaster risk analysis in the Philippines due to cyclones.

102
The researcher in this study is the focus to improve the output of the hybrid model. In many cases, ANN is providing better prediction accuracy. Japan is the very venerable 116 geographical area for floods, which facing an increase in water level after Typhoon. In 117 this paper, the author used various datasets. This work aims to select the most relevant 118 dataset for the ANN-based water flow forecast model [Kim et al. 2016]. The study area in 119 this paper is the Pampanga river basin. The hardware setup contains a microcontroller, 120 solar penal, ultrasonic sensor, and GSM module. A Feedforward network with 121 backpropagation is used and for optimizing the network Levenberg-Marquardt training 122 algorithm is used [Sahagun et al. 2017]. In some cases, a tree-based ML model is also 123 used to predict the sensitivity of flooded areas based on the spatial parameters [Lee et al.           system. We demonstrate the working of our proposed methodology. The process is 286 started from collect data from the data storage center then we pre-process the initial data.

306
In our model, to introduce non-linearity into the model ReLU activation function is used.

307
For classification, the softmax activation function is used to obtain the probability of 308 being between 0 and 1. The equation of ReLU and softmax function is given below.

309
In neural networks, the Cross-entropy loss function is used while training our network.  High precision requires a low FN rate and a low FP rate.  we discuss the graphical representation and identify that which machine learning     can conclude that our proposed model gives more accuracy than other models used for 515 prediction. In the future, if we use more environmental parameters the there is a chance to 516 achieve higher accuracy. Also if we combine other models and develop a hybrid model 517 for flood prediction then also we get better accuracy.

519
Compliance with ethical standards 520 521 Conflict of interest There is "NO any" direct or indirectly related conflict of interest for 522 this manuscript and with authors.