Implementation Self Organizing Map for Cluster Flood Disaster Risk

Indi Febriana Suhriani, Lalu Mutawalli, Baiq Rina Ari Widiami and Chumairoh

Universitas Gadjah Mada, Indonesia

Keywords: Cluster, Flood Disaster, SOM.

Abstract: Floods have devastating effects on human, economic, and environmental life. The flood risk can't be avoided

completely so it must be managed. Flood disaster management does not seek to eliminate the danger of

flooding, but to cope with it. Thus, this study aims to (1) classify the provinces in Indonesia based on the

results of flood risk analysis; (2) Identify flood risk characteristics in each group; and (3) Analysist the flood

risk level of each province in Indonesia. The research method used in this research is the method of Self

Organizing Map (SOM) by using software R. This research conducted cluster class based on flood risk

variables, i.e. province, number of incidents, a victim, houses and damages. The results showed the grouping

divided into 6 clusters with members of each cluster are; cluster 1 (central of java and east java), cluster 2

(west java), cluster 3 (Aceh, North Sumatra, West Sumatera, Riau, Jambi, South Sumatra, Bengkulu,

Lampung, Bangka Belitung, Kepulauan Riau, DKI Jakarta, Bali, Nusa Tenggara Barat, East Kalimantan,

North Sulawesi and Papua), cluster 4 (Banten, West Papua), cluster 5 (South Kalimantan), and cluster 6

(central).

1 INTRODUCTION

Indonesia is one country that has disaster-prone areas,

one of the disasters that often befall Indonesia is

flooded. Floods have an impact that can be bad for

human life. Flood disaster can affect the disruption of

the economy and the environment (UNISDR. 2015).

Negative impacts caused by floods are important for

anticipatory steps. To take anticipatory steps requires

knowledge of the hazards of their impact (Zischg, et

al., 2018). There have been several previous studies

on the analysis of the negative impacts of floods.

Studies show that the impact of flooding is not only

on aspects of buildings and materials. However, the

effect on the prevalence of increased psychological

illnesses after the flood disaster (Zhong, et al. 2018).

To explore the linearity of relationships using

spatial analysis and temporal variants of the impact of

floods. The results show that most and most affected

are nonlinearly properties especially those close to the

river (Rajapaksa, Zhu, & Lee, 2017). Hydrographic

construction design for the study of climate change

impacts, informative hydrograph development in

flood disaster impact studies (Brunner, Sikoska, &

Seibert, 2018). Factor analysis that influences flood

by using weighted overlay technique approach,

research result can give information about danger

zone of food for early (Azmeri, Hadihardja, & Vadia,

2016)

Several previous studies have conducted various

analysis approaches to the impact of floods. In this

study we propose anticipatory steps in obtaining flood

information in the territory of Indonesia. This

research can provide knowledge by conducting flood

risk level analysis at every province in Indonesia. In

order for prevention and mitigation measures to be

carried out properly, effectively and efficiently.

The method proposed in this research is the Self

Organizing Map method (SOM) and the aid of

computation method using R programming language

to support the faster analysis process done. The self-

organizing map is a statistical data analysis method of

the branch of unsupervised learning, whose goal is to

determine the properties of input data without explicit

feedback from a teacher (Martin & Obermayer,

2009). The SOM algorithm creates mappings which

transform high-dimensional data space into low-

dimensional space in such a way that the topological

relations of the input patterns are preserved (Köküer

& Green, 2007). Some previous studies used the self-

organizing map method. The use of SOM integrated

with image processing to model the detection of

damaged gaps in bridges (Chen, et.al, 2017). SOM

implementation can be used to model traffic

Suhriani, I., Mutawalli, L., Widiami, B. and Chumairoh, .

Implementation Self Organizing Map for Cluster Flood Disaster Risk.

DOI: 10.5220/0008522604050409

In Proceedings of the International Conference on Mathematics and Islam (ICMIs 2018), pages 405-409

ISBN: 978-989-758-407-7

405

disruption patterns (Steiger, et.al., 2016). SOM

implementation can effectively help better rescue

planning in the aftermath of a disaster. The built

model is used to produce a risk map of survival. Some

problems can be modeled and, in the map, using the

SOM approach. The SOM method adopted in this

study to cluster can provide clustering of areas most

affected by floods. This study aims to provide early

knowledge about the potential for flood disasters in

the territory of Indonesia.

2 RELATED WORK

2.1 Self Organizing Map

Kohonen Self Organizing Maps is a network found

by Teuvo Kohonen is one of the most widely used

networks. Named "self-organizing" because this

method does not require a special surveillance and

SOM approaches through unsupervised competitive

experiments. The word "maps" it self because this

method using the map in weighting input data. Each

node in this network works to present each data input,

therefore this network can also be called "Self-

Organizing Feature Maps", the concept of "features"

becomes an important and valuable thing, specifically

the topology relationship between the inputted data

will be maintained and original when mapped in the

SOM network (Guthikonda, 2005).

In perspective, SOM can be seen not only as a

tool but as a toolbox that contains the features of

numbers and can be more interesting in different

situations. SOM Kohonen network has three topology

types, namely linear array, rectangular, and

hexagonal. Linear array topology shows cluster units

arranged linearly

Figure 1: Kohonen Topological Map.

Source: Yusob, B., et al., (2013)

2.2 Cluster Validation

Cluster validation is a procedure that evaluates the

results of quantitative and objective cluster analysis.

Index Dunn is one part of the cluster validity such as

the Davies-Bouldin index or the Silhouette index in

this case Dunn index is an internal evaluation scheme

where the result is an internal evaluation scheme.

  









(1)

min

= smallest distance between observations on

different clusters

max

= largest distance in each data cluster

3 RESULTS AND DISCUSSION

Based on the objective of this research is to group the

provinces in Indonesia and identify the characteristics

based on flood disaster risk in 2008-2018 by using

Self Organizing Maps method. The Kohonen network

is used to divide input patterns into clusters. Suppose

the input is a vector consisting of n components to be

grouped within a maximum of group m. SOM

network requires a training progress to minimize the

average distance of an object to the nearest.

In figure 2 below is below is the number of

progress training that shows the number of iterations

against the average distance to the nearest unit.

Progress or iteration training is used to find out how

much it takes for the cluster to be optimal, the more

iterations are carried out, the smaller the distance of

cluster units and the results of grouping will be better.

After passing an iteration of approximately 4000

shows that the progress of the training begins to

stabilize with the average cluster unit distance below

0.05 and the researcher uses 10,000 iterations to do

this grouping. To help researchers determine which

clusters produced in SOM can be done using the

Cluster Amount of Squares (WCSS). The Cluster

Sum of Squares (WCSS) is able to help groups

desired by researchers.

ICMIs 2018 - International Conference on Mathematics and Islam

406

Figure 2: Training Process.

Figure 3. Seen at the time of forming one cluster,

curve still shows steepness and increasingly sloping

according to cluster increase. In essence the number

of clusters that are formed when the number of

members who joined fewer and fewer. To prove

whether the number of clusters selected is correct or

has not been used cluster validity test, the following

is the output of programs R on the test of cluster

validity:

Figure 3: Within Cluster Sum of Squares.

Figure 4 cluster validation can be known if the

cluster to be formed is 4 clusters up to 8 cluster then

the best cluster is 6 clusters, as shown in the picture

above optimal scores shows there are two methods

that choose the 6 most reverse clusters are Dunn and

Silhouette, while connectivity shows that 4 clusters

are the best number of clusters, because there are 2

that say that 6 clusters are the best result, the

researcher will use 6 clusters for flood disaster

grouping in Indonesia by province using 6 clusters.

After the number of clusters is determined then the

next process is to make fan diagram.

Figure 4: Cluster Validity Test.

The results of the provincial grouping using SOM

are as shown in table 1 below.

Table 1: Number and Grouping Members.

Group

Number

Member

Members of the group

Central Java, East Java

West Java

Aceh, North Sumatera, West

Sumatera, Riau, Jambi, South

Sumatera, Bengkulu,

Lampung. Bangka Belitung

Islands, Riau Islands, DKI

Jakarta, Yogyakarta, Bali,

West Nusa Tenggara, East

Nusa Tenggara, West

Kalimantan, East Kalimantan,

North Sulawesi, North

Sulawesi, Central Sulawesi,

South Sulawesi, Southeast

Sulawesi, Gorontalo, West

Sulawesi, Maluku, North

Maluku, Papua

Banten, West Papua

South Kalimantan

Central Kalimantan

The mapping results from the grouping analysis

using Self Organizing Maps are shown in Figure 5.

Figure 5: Mapping Using SOM Algorithm.

Implementation Self Organizing Map for Cluster Flood Disaster Risk

407

SOM grouping table and SOM mapping visually

and see the table of group average, cluster 1

consisting of Central Java and East Java has the

largest number of occurrences. Successively 929 and

963 incidents with the number of house damage

(heavily damaged, medium, light, submerged) also

have a high value. This group corresponds to the one

associated with the blue circle in Self Organizing

Maps.

West Java in an orange circle is cluster 2 in Self

Organizing Maps output has a high number of

occurrences with damage of house (damaged heavy,

medium, light, submerged) which is most numerous

than a group. Victims (died, injured, suffered) are

high while the number of injured is very large that is

37195 inhabitants, higher than other groups. The

green circle is associated with 27 provinces: Aceh,

North Sumatera, West Sumatera, Riau, Jambi, South

Sumatera, Bengkulu, Lampung, Bangka Belitung

Islands, Riau Islands, Jakarta, Yogyakarta, Bali, West

Nusa Tenggara, East Nusa Tenggara, West

Kalimantan, East Kalimantan, North Kalimantan,

North Sulawesi, Central Sulawesi, South Sulawesi,

Southeast Sulawesi, Gorontalo, West Sulawesi,

Maluku, North Maluku, Papua are incorporated in

cluster 3 has an average.

The number of occurrences with the category of

damage to the house (heavily damaged, medium,

light, submerged) is very small, so it has a low

average value too. On average the number of victims

(died and missing, injured, suffered and displaced)

and also broke facilities (health, worship, education).

Banten and West Papua are associated in a red circle

on Self Organizing Maps, which is cluster 4, where

this cluster has the lowest number of events compared

to other clusters.

Table 2: Cluster Profile.

Cluster 1

Cluster 2

Cluster 3

Cluster 4

Cluster 5

Cluster 6

Number of events

946

786

126.96

85.5

184

116

People Died and

Missing

112

148

28.93

250

Injuries

14731.5

37195

532.96

8351

381

860

Suffered and

displaced

996364.5

2013938

257086.8

319338.5

4860178

398342

Heavy Damage

House

1791

4839

759.22

966

262

Houses Damaged

Medium

1954

2455

130.3

Light Damaged

House

6928

9232

2776.81

179.5

275

3349

House Submerged

393278

647737

51033.3

48395

235426

58527

Health facility

0.5

0.78

Worship Facilities

115

4.44

131

Educational Facilities

29.5

10.63

158

The low number of damages to houses (severely

damaged, medium, light, submerged) and poor

facilities (health, worship, education) are still needed

for education or training on disaster mitigation due to

the average. The highest number of deaths and

disappearances compared to other clusters, while the

number of victims (injured, suffered and displaced)

has a moderate average. In cluster 5, South

Kalimantan is associated with a purple circle, for this

cluster has the lowest number of events, and the low

number of damage to the house (damaged, medium,

light) is accompanied by low facility damage (health,

worship, education), but the number of houses soaked

in water is so high that the number of displaced

persons also has the highest number among other

clusters. Finally, central Borneo is associated in a

brown circle with a low number of incidents, and a

low number of (low, medium, light, submerged)

damage to homes, and low casualties (dead and

missing, injured, suffered and displaced) also, in this

case need improvements in terms of facilities (health,

worship, education) because it has the highest

damage value compared with other clusters.

4 CONCLUSIONS

The number of groups formed as many as 6 groups, is the

number determined by researchers with the approach using

Within Cluster Sum of Squares. The use of Self Organizing

Maps algorithm resulted in a grouping with a group of 27

provinces, two groups of 2 provinces and three groups of 1

province each. Groups 1 and 4 each consist of Central Java,

East Java and Banten, West Papua. Groups of 2.5.6 and 4

ICMIs 2018 - International Conference on Mathematics and Islam

408

each consist of West Java, Central Kalimantan, and South

Kalimantan. Group 3 consists of Aceh, North Sumatera,

West Sumatera, Riau, Jambi, South Sumatra, Bengkulu,

Lampung, Kep. Bangka Belitung, Kep. Riau, DKI Jakarta,

In Yogyakarta, Bali, West Nusa Tenggara, East Nusa

Tenggara, West Kalimantan, East Kalimantan, North

Sulawesi, North Sulawesi, Central Sulawesi, South

Sulawesi, Southeast Sulawesi, Gorontalo, West Sulawesi,

Maluku, North Maluku, Papua.

ACKNOWLEDGEMENTS

Thanks to the Islamic University of Indonesia which has

provided budget support for the implementation of the

research process and budget to attend the international

conference which will be held by the Universitas Islam

Negeri Mataram West Nusa Tenggara.

REFERENCES

Azmeri, Hadihardja, I., & Vadia, R. 2016. Identification of

Flash Flood Hazard Zones in Mountainous Small

Watershed of Aceh Besar Regency Aceh Province

Indonesia. The Egyptian Journal of Remote Sensing

and Space Science, Vol 19(1), 143-160.

Brunner, M. I., Sikoska, A., & Seibert, J. 2018. Bivariate

Analisis of Flood in Climate Impact Assessments.

Science of The Total Environement, 1392-1403.

Chen, J. H., Su, M. C., Cao, R., Hsu, S. C., & Lu, J. C. 2017.

A Self Organizing Map Optimization Based Image

Recognition and Processing Model for Bridge Crack

Inspection. Automation in Construction, Vol 73, 58-66.

Guthikonda, S. M. 2005. Self-Organizing Maps. In

Kohonen, & Teuvo, Self-Organizing Maps (p. 10). US:

Springer.

Köküer, M., & Green, R. 2007. Towards Automatic Risk

Analysis for Hereditary Non-Polyposis Colorectal

Cancer Based on Pedigree Data. Outcome Prediction in

Cancer, 319-337.

Martin, R., & Obermayer, K. 2009. Self-Organizing Maps.

Encyclopedia of Neuroscience, 551-560. Academic

Press.

Rajapaksa, D., Zhu, M., & Lee, B. 2017. The Impact of

Flood Dynamics on Property Values. Land Use Policy,

Vol 69, 317-325.

Steiger, E., Resch, B., Albuquerque, J. P., & Zipf, A. 2016.

Mining and Correlating Traffic Events from Human

Sensor Observations with Official Transport Data

Using Self Organizing Maps. Transportation Research

Part C: Emerging Technologies, 73, 91-104.

UNISDR. 2015. Making Development Sustainable: The

Future of Disaster Risk Management. United Nations:

Geneva.

Yusob, B., et al. 2013. Spiking Self-Organizing Maps for

Classification Problem. Technology Came, Vol 11, 57-

64.

Zhong, S., Yang, L., Toloo, S., Tong, S., Sun, X.,

Crompton, D., et al. 2018. The Long Term Physical and

Pysicological Health Impacts of Flooding: A

Systematic Mapping. Science of The Total

Environment, 165-194.

Zischg, A. P., Hofer, P., Mosimann, M., Röthlisbergerab,

V., Ramirez, J., Keiler, M., et al. 2018. Flood Risk (d)

Evolution: Disentangling Key Driver of Flood Risk

Change with a Retro Model Experiment. Science of The

Total Environment, 195-207.

Implementation Self Organizing Map for Cluster Flood Disaster Risk

409