Disaster Impact Mitigation using KDD and Support Vector Machine algorithms

Disasters such as Hurricanes, Typhoons, Floods and earthquakesare not good for the society since it causes serious damage for the society. A natural disaster causes loss in property as well as in life of victims. The victims need immediate help once they are affected by the disaster. The immediate need are rescue, food and communications. The survey says victims of recent disaster were unable to get instant communication regarding the evacuation path and other help from authorities for remedial action. This can be overcome with our proposed idea of having a database of area wise population along with the pre-disaster and post-disaster satellite images of the disaster affected area. Knowledge Discovery in Databases (KDD) is used in data pre-processing and to extract knowledge from the database. Support vector machine(SVM) is used to classify the disaster effect with the pre-disaster and post-disaster satellite images as input. The idea is implemented and tested with sample data and has given impressive results


Introduction
A natural disaster can causes loss of life as well as property damage.Top ten deadliest natural disasters include Calcutta cyclone in 1737 with death toll of 300000, India Cyclone in 1839 with death toll of 300000 and Indian ocean earth quake & Tsunami in 2004 with overall death toll of 280000.Relief involves moving and keeping emergency equipment; shifting people out of danger; supplying required food, shelter and medical services.In this paper, we propose an idea of predicting the impact of disaster within a few minutes of its occurrence.This will help in the speedy recovery and help in the evacuation of people to a safer place.The paper is organized as follows.Section II gives the complete details about review status of research and development in this subject.Section III illustrates the proposed methodology using KDD and SVM.The performance evaluation of proposed design is illustrated in section IV.Section V concludes the paper.

Review Status Of Disaster Management
The review of literature was done in three different areas.
• The review on Pre-processing the data and representing them  The review on predicting the impact of disaster using satellite map

•
The review on communicating the victims during disaster within a short period Pre-processing the data and representing them is explained by Igor Kononenko et al., [1].Datta et al., have used feature selection algorithm for pre-processing.Here different features are selected and they are used to improve the accuracy of the classifier.They have used the classifier such as Naïve Bayes, HNB and NB Tree.[2] Wei Zhao et al., claims that change detection method could be used for landslide mapping and they have tested with post-earth quake images.They have proposed a pixel based change detection method using the satellite image from Landsat-8 and have tested using 2015 Nepal Earth quake images [3].Siti Nor Khuzaimah et al., have used neural network with three convolutional layers for disaster detection.They have tested by creating disaster detection training data patches and have shown an accuracy of 80 -90 % [4] Kamran Ali et al., proposed a disaster management communication networks.They have proposed a network architecture design.They have used the existing network infrastructure and have integrated with layer-based techniques.They have used cloud processing concepts.[5].
Flor Alvarez et al., have enumerated the risks and challenges that may arise during crisis and have provided practical guidelines for mitigating them during crises.[6] Apart from these the social media can also be used for impact prediction as proposed by Soumya et.al [7].They have made a new event detection approach my combining the linguistic feature of twitter.The traffic and its effect during disaster is proposed by Lin Qui et.al [8].Here a parallel shortest path algorithm in road network is given

Impact Prediction Using KDD and SVM
The input data are details about the population, houses, hospitals, senior citizens, infants and Geographical map.The data is pre-processed using KDD algorithm [9], [10], [11].It is shown in Fig 1 .In the Understanding phase, the data is analysed and represented in a tabular form.In the preparation phase, sampling and data cleaning operations are performed.This includes checking the completeness of data records, correcting or removing for noise and missing values, etc.In the next phase the knowledge is discovered from the cleaned data.This knowledge includes the % of people in a particular area at any period of time which would be affected by the employment, climate vacation period etc.For eg., It is expected that the impact of disaster on life of people will be more during might time than day time.
The first level output is a map when zoomed will give the details of the selected area.A sample is shown in Fig 2 In the Understanding phase, the data is analyzed and represented in a tabular form.In the preparation phase, sampling and data cleaning operations are performed.This includes checking the completeness of data records, correcting or removing for noise and missing values, etc.In the next phase the knowledge is discovered from the cleaned data.This knowledge includes the % of people in a particular area at any period of time which would be affected by the employment, climate vacation period etc.For eg., It is expected that the impact of disaster on life of people will be more during might time than day time.Let Samples be X={(x-1,y-1),(x-2,y-2),…(x-n,y-n)} where xi is element of R M , yi lies between -1 and +1.The hyperplane is w T x+b=0 with normal vector w and bias b in order to separate two margins.

Evaluation of
The margin is calculated using equation ( 1) The output of this will have two information. 1.
The impact of the disaster 2.
The rescue path for the people The disaster impact predicted could be used by the Government officials and NGOs for helping the needy The rescue path can be used by the people in any area to save themselves.
The overall block diagram of disaster prediction and management is shown in Fig 3.

Input the factors and values
New map after disaster with impact predicted and rescue information Support Vector Machine

Evaluation
The evaluation of the work is done in three phases.
• The accuracy of the knowledge generated by KDD • The accuracy of SVM in disaster impact prediction • The accuracy of the algorithm in finding evacuation path The knowledge represented by KDD gave accuracy of 95% since the data input is more cleat.
The verification is done with Earth quake and Flood data.
Experimental setup for this work is done as follows • Earth quake -20 train data sets (200 patches per set) -10 test data sets (200 patches per set) • Flood experiment (Tsunami)-5 train data sets (100 patches per set) -3 test data sets (100 patches per set) • For each set of data, we have minimum 2 aerial images which includes the input and output images: Sample input is shown in Figure 4. [12] • The impact predicted was compared with the exact impact given by the government.It gave around 80 -90 % accuracy.
The rescue path was tested with test data of earth quake and Flood images.The earth quake test data gave accuracy of 98 % and the flood data gave accuracy of 80%

Conclusion and Future Work
The result could be tested with few real examples only.The data set used were 10% real data and 90% generated data.Experimental evaluation gave a result around 90% overall.
The final product could be used as an application that can be downloaded by any user.Especially by the residents of disaster prone area.
When there is no power and no internet connection the work may not be useful for the residents.This could be overcome by placing wireless FM transmitter transmitting rescue information in a balloon and made fly over affected area.
The algorithm will not be at its best during night time and flood times since the map may not be clear.This can be taken as future work by receiving the immediate inputs from the residents using mobile or any other device and hence predicting the disaster impact as well as evacuation path.

Fig 4 :
Fig 4: Pre and Post tsunami pictures