=Paper=
{{Paper
|id=Vol-2247/poster3
|storemode=property
|title=Strategy for Generation of Knowledge through Automatic Correlation of Dimensional Data in Star Schema: Application in the Context of Leishmaniasis
|pdfUrl=https://ceur-ws.org/Vol-2247/poster3.pdf
|volume=Vol-2247
|authors=Wallace Pinheiro,Geraldo Xexeo,Jono Souza,Ana Bárbara Pinheiro,Ciro Gomes
|dblpUrl=https://dblp.org/rec/conf/vldb/PinheiroXSPG18
}}
==Strategy for Generation of Knowledge through Automatic Correlation of Dimensional Data in Star Schema: Application in the Context of Leishmaniasis ==
https://ceur-ws.org/Vol-2247/poster3.pdf
Strategy for Generation of Knowledge through
Automatic Correlation of Dimensional Data in Star
Schema: Application in the Context of Leishmaniasis
Wallace Anacleto Pinheiro1,2[0000-0001-7076-8785], Geraldo Xexéo1[0000-0003-3975-9076], Jano
Moreira de Souza1[0000-0001-5080-1955], Ana Bárbara Sapienza Pinheiro3[0000-0002-2766-3141],
Ciro Gomes3[0000-0002-3069-6884]
1 COPPE – Graduate School and Research in Engineering – Universidade Federal do Rio de
Janeiro (UFRJ) – Rio de Janeiro, Brazil
2IME – Instituto Militar de Engenharia – Rio de Janeiro, Brazil
UNB – Universidade de Brasília, Brasília, Brazil
wallaceapinheiro@gmail.com
xexeo@cos.ufrj.br
jano@cos.ufrj.br
absapienza@gmail.com
ciromgomes@gmail.com
Abstract. The aim of this article is to propose a methodology for generation of
knowledge through correlation techniques, based on dimensional data obeying a
star model. This methodology relates and integrates data from one or more the-
matic areas (data marts) using correlation coefficients applied to data derived
from facts and dimensions of a data mart. In order to analyze the application of
the strategy, we selected the database of Brazil Information System for Notifiable
Diseases (SINAN) of Brazilian Department of Health to gather information about
a particular neglected disease that affects several Brazilian capitals: Visceral
Leishmaniasis. The proposed methodology has generated several correlations
that can support the planning of public strategies to combat this disease.
Keywords: Correlation, Star Schema, Leishmaniasis, Neglected Diseases.
1 Introduction
Media in Brazil frequently disclose information on outbreaks of diseases in places that
were not previously endemic [1]. Trade, means of transportation, foreigners without
immunity to local pathologies, residents without immunities to foreign pathologies are
elements that mix, narrow borders, break geographic boundaries, connect cities and
diseases as well.
Data on climate, characteristics of the inhabitants of a region and characteristics of
the place where they live can provide valuable information to understand and plan ac-
tions that seek to prevent or minimize the outbreak of diseases that spread among vari-
ous cities in the world.
2
Thus, this article proposes the use of correlation techniques on data related to a ne-
glected disease that affects several Brazilian cities and municipalities: Visceral Leish-
maniasis. From 2007 to 2015, in Brazil, the annual average number of reported cases
related to all types of leishmaniasis exceeded 20,000 notifications, of which 4,000 no-
tifications reported visceral leishmaniasis. According to some authors [2], there is an
increasing urbanization of some types of Leishmaniasis in certain Brazilian cities. In
the case of a vector-borne parasite (ex.: sandflies), some indicators may be essential for
the survival of the parasite and the vector considering the environment. The vector, in
particular, requires the presence of some freshwater source (rivers, lakes, etc), having
an area of action of only some hundreds of meters from these sources, due to its small
size. It is a hematophobic insect, also known as sandfly. With this knowledge, some
authors [3] report that increased temperature and rainfall precipitation contribute to the
appearance of new cases of leishmaniasis. Other authors [4] believe that the dry climate
and low altitudes contribute to the presence of the disease, since they benefit the pres-
ence of the vector.
Therefore, considering the increasing acting range of vector, and that the disease
occurs in different clinical forms, from the tegumentary to the visceral, several cities in
the American continent, especially some Brazilian cities, are vulnerable to this growing
epidemiological problem [5]. Although there are different types of leishmaniasis, de-
pending on the parasite and the immunity of the individual, the process of contagion is,
basically, the same. Deforestation, climate change, the reservoir with several hosts, the
peri-urban area, ecological tourism, among others, are some risk factors. It is estimated
that there are two million new cases of leishmaniasis in the world each year [6], mainly
in third world countries.
This work proposes that the notification data on Visceral Leishmaniasis stored in
Brazil Information System for Notifiable Diseases (SINAN) of Brazilian Department
of Health be correlated, seeking to relate characteristics of the affected population with
the increase or decrease in the number of cases. It also proposes to correlate data from
SINAN with data on climate provided by Brazilian Institute of Research and Forest
Studies (IPEF). Thus, we intend to obtain an overview of possible links between factors
that may contribute to the increase or reduction of the number of cases of leishmaniasis
according to each region.
In this work, we do not discuss the use of causality techniques, we only analyze
correlations between data and their coherence.
The rest of this work is organized as follows: Section 2 presents the related works,
Section 3 presents the experiments configuration, Section 4 presents results and Section
5 shows the final considerations and the proposals for future work.
2 Related Work
SINAN is a system provided by Brazilian Department of Health that stores data on
compulsory notification diseases or injuries in Brazil [7]. It provides historical data,
allowing them to be analyzed on different aspects. Some works use SINAN data to
relate diseases or injuries with climatic factors or characteristics of a region or work
3
[3], [8]. However, these studies use specific models to relate data in specific regions,
not using correlation factors for a broader analysis.
In the area of information technology, some researchers [9] [10] cite data mining
techniques that can be used in different types of databases, including data marts and
data warehouses, such as trees Bayesian classification, association rules (correlation),
neural networks, clustering, and genetic algorithms.
There are studies aimed at correlating datasets that represent multidimensional dis-
tributions using correlation coefficients, such as Pearson coefficient [11]. Correlation
is done using characteristics of the distribution, such as linearity and non-linearity.
These works are not related to dimensional modeling used in a data warehouse (dimen-
sion and fact), but addresses multi-dimensional data distributions. However, we did not
find in literature works that perform correlation formed by combining a dimension of a
star schema with time dimension from fact metrics.
3 Experiment Configuration
The approach proposed in this work can be applied in different areas of knowledge,
since information be stored in a star schema based repository. Thus, we decide to apply
it in the context of tropical neglected diseases and climate, a relevant topic, since the
world is passing through many climate changes, while some tropical neglected diseases
are spreading in new regions [2], [4], [5].
This work materializes the proposed strategy using data from SINAN of Brazilian
Department of Health, available at: http://portalsinan.saude.gov.br/, and data from Bra-
zilian Institute of Research and Forestry Studies (IPEF), available on the website:
http://www.ipef.br/geodatabase/. Both SINAN and IPEF provide their data in file for-
mat. It was chosen to extract data on a specific neglected tropical disease: Visceral
Leishmaniasis.
Leishmaniasis data, extracted from SINAN, were combined with climatic data on
Brazilian municipalities, extracted from IPEF, in a single data repository with two data
marts that served as basis for correlations.
Fig. 1 presents an overview of the process executed for creation of the dimensional
database, and the element highlighted in gray corresponds to the components of the
solution proposed in this work.
In this context, we executed the task of obtaining the data through an ETL process,
using Pentaho Data Integration tool, Community version [12]. Extraction-Transfor-
mation-Load (ETL) tools, such as Pentaho, are widely used to integrate data from het-
erogeneous sources.
After passing through the ETL process, Data Correlation and Filtering module cal-
culated the correlations. The implementation used Python 3.6 language and MicroStrat-
egy Desktop 10.8.0 tool.
4
Fig. 1. – Proposed Architecture
Once stored in the database following the specification of a star schema, the pro-
posed solution processed data in order to provide the correlations between the modeled
dimensions. The number of processed data is showed in Table 1. It is possible to notice
that few registers from the queries generate a large number of correlations.
Table 1. Quantity of Processed Data
Description of the Result Quantity
Number of Registers involving Climate Fact (IPEF) 1500
Number of Registers involving Notification Fact (SINAN) 6590
Number of Correlations - Intra SQLs (Climate Fact) 1500
Number of Correlations - Intra SQLs (Notification Fact) 87740
Number of Correlations - Between SQLs (Climate Fact) 3750
Number of Correlations - Between SQLs (Notification Fact) 82496
Number of Correlations - Between Facts 73000
4 Results
Using the implementation discussed in previous sections, we have processed data stored
on Leishmaniasis considering the period from January 2016 to December 2016 of Bra-
zilian capitals. We considered only correlations greater than 50% (direct or inverse),
having a significance level of at least 1%.
Fig. 2 shows the correlations found during this process. The label of the nodes iden-
tifies the correlated data, the size of the nodes as well as the width of the arcs are related
to the number of correlations of the node, the color of the arcs represents the intensity
of the correlation (the greater the intensity of the red, closer to the lower value of the
correlation filter; while this, the greater the green intensity, the closer the upper value
of the correlation value). Arrows indicate the direction of correlation. Names in the
graph correspond to Brazilian capitals. Numbers ranges are related to age groups. Code
B550 corresponds to the International Code of Diseases (ICD) for Visceral Leishman-
iasis (total cases). Some details of these correlations are discussed in the next para-
graphs.
5
Fig. 2. – Correlations
Fig. 3, which details some correlations presented previously about temperature, rainfall
index and number of cases among Brazilian capitals (these data are from two different
data marts). Considering temperature, the highest value occurs between Palmas and
Fortaleza (value of approximately 86%). When analyzing the rainfall indexes, the high-
est correlation values are obtained for Fortaleza and Aracaju (48%). Considering the
relationship between the case number, it is observed that Fortaleza and Palmas are cor-
related (value slightly above 50%).
It is possible to observe that correlations 1 and correlations 3 contains Fortaleza and
Palmas. These results suggest that, in the case of these two capitals, the temperature
variation influences more the variation of the number of cases than the rainfall index.
6
Fig. 3. Correlations for Temperature (Correlations 1), Correlations for RainFall Index (Correla-
tions 2) and Correlations for Number of Cases (Correlations 3)
However, if we look at the other spectrum of correlation, that is, the inverse correla-
tions, we can see that there is an inverse correlation higher than 70% between the num-
ber of cases in Palmas and the rainfall index of several capitals, including Palmas, ob-
served in the self-relationship, as shown in Fig. 4.
This shows that in times of less rainfall, the number of cases in Palmas tends to
increase. As seen earlier, higher temperatures also increase the number of cases in this
region, providing a picture of the most critical periods in which vector-control measures
should be taken. This, in a first moment, is not in agreement with the view presented
by certain authors [3] that with increasing humidity there is an increase in the number
of cases, but corroborates the view presented by others [4] who believe that Leishman-
iasis is related to dry climates. In this scenario, the great advantage of the proposed
strategy is to be able to base its observations on a large number of data from all notifi-
cations of Visceral Leishmaniases cases, considering Brazilian territory and not just
local information.
Fig. 4. – Inverse correlations considering rainfall index
7
Another possible analysis involves data from two different tables of the data mart for
notifications, it is verified that there is a correlation higher than 50% between cases of
leishmaniasis in Fortaleza for 60 to 64 age range, shown in Fig. 5. This can guide the
planning of public policies to combat this disease in this place. Thus, it is possible to
observe that, from a visual analysis of the correlation graph, several information can be
obtained, providing information for potential assertive strategies by decision makers.
Fig. 5. Correlations between Fortaleza and Age Range
5 Final Considerations
Correlations between dimensions or their data can aid in generating new insights as
they provide measurements of relationships over data marts that normally store a large
amount of data.
The analysis of correlations of Brazilian capitals brought some relevant information:
in some regions there was a direct relation between temperature and number of cases
and an inverse relationship between rainfall index and number of cases. It is also pos-
sible to associate variation of the number of cases with specific age groups. These data
could be used to guide public policies in Brazilian capitals, aiming to support the epi-
demiological control of leishmaniasis.
It is worth mentioning that previous studies on correlations involving Leishmaniasis
have focused on specific sites, evaluating factors such as: climate (for example: rainfall
and temperature) and population characteristics. However, a broader view that corre-
lates a large volume of data provides a new light on this subject, since temperature,
rainfall index and population characteristics are only some of the factors that may con-
tribute to increase the number of cases of this disease.
Therefore, future studies will add new information from data sources considering
other neglected diseases. The objective will be to analyze, in addition to correlations
between the factors presented in this study, the effect of coinfections on evolution of
diseases; in particular those transmitted by vectors.
8
References
1. Oliveira, T. Leishmaniose atinge mais municípios e avança em direção à capital (in portu-
guese). Veja - São Paulo (2018).
2. Bevilacqua, P. D., Paixão, H. H., Modena, C. M. & Castro, M. C. P. S. Urbanization of vis-
ceral leishmaniose in Belo Horizonte, Brazil. Arq. Bras. Med. Veterinária e Zootec. 53, 1–8
(2001).
3. Mendes, C. S., Coelho, A. B., Féres, J. G., Souza, E. C. de & Cunha, D. A. da. Impacto das
mudanças climáticas sobre a leishmaniose no Brasil (in portuguese). Cien. Saude Colet. 21, 263–
272 (2016).
4. Pinto, I. de S., Santos, C. B. dos, Grimaldi Jr., G., Ferreira, A. L. & Falqueto, A. American
visceral leishmaniasis dissociated from Lutzomyia longipalpis (Diptera, Psychodidae) in the
State of Espírito Santo, Brazil. Cad. Saude Publica 26, 365–372 (2010).
5. CDC. CDC - Leishmaniasis - Disease. Centers for Disease Control and Prevention (2013).
6. MS. Manual de vigilância da leishmaniose tegumentar (in portuguese). Ministério da Saúde
(2017).
7. Barbosa, D. A. & Barbosa, A. M. F. Avaliação da completitude e consistência do banco de
dados das hepatites virais no estado de Pernambuco, Brasil, no período de 2007 a 2010 (in por-
tuguese). Epidemiol. e Serviços Saúde 22, 49–58 (2013).
8. Albuquerque, P. C. C. de et al. Health information systems and pesticide poisoning at Per-
nambuco. Rev. Bras. Epidemiol. 18, 666–678 (2015).
9. Fayyad, U. M., Piatetsky-Shapiro, G., Smyth, P. & Uthurusamy, R. Advances in knowledge
discovery and data mining. (AAAI Press, 1996).
10.Han, J., Kamber, M. & Pei, J. Data mining: concepts and techniques. (Elsevier Science, 2011).
11.Zhang, Y., Liu, T., Li, K. & Zhang, J. Improved visual correlation analysis for multidimen-
sional data. J. Vis. Lang. Comput. 41, 121–132 (2017).
12. M., V., Syed, A., Mohammad, A. & N., M. Pentaho and Jaspersoft: A Comparative Study of
Business Intelligence Open Source Tools Processing Big Data to Evaluate Performances. Int. J.
Adv. Comput. Sci. Appl. 7, (2016).