Urban Management: Learning from Green Infrastructure …
507
4 Conclusions: Classification Models Support Planning
and Public Policies for Urban Environment Towards
Sustainability
This study demonstrated the potential of data mining techniques to extract information of fairly easily available municipal indicators, such as demographic density (DD), Gross Domestic Product (GDP), Municipal Human Development Index
(M_HDI), percentages of households in streets with adequate sanitation (SNT), with
trees (ARB) and with urbanization (URB), and also the number of vehicles (VHCLS)
and morbidities of respiratory (RD) and circulatory (CD) diseases, responsible for
high rates of causes of death and hospitalizations in the State of Paraná and in Brazil.
This can help establish better urban management practices and reinforce some public
policies.
The variable that showed the largest variability was DD and the one with the
least variability was M_HDI. Most of the municipalities in the State of Paraná are
classified as small, with up to 50,000 inhabitants, and 70% (about 280 of 399) of the
municipalities presented DD equal to 3 to about 35 inhabitants per square kilometer. Meanwhile, the density of the largest cities in Paraná—Londrina, Maringá and
Curitiba (main city)—was 306.52, 733.14 and 4027.04 inhabitants per square kilometer, respectively; whereas Rio de Janeiro and São Paulo, the largest in the country,
counted 5265.82 and 7398.26 inhabitants per square kilometer, respectively, in the
same census of 2010. This revealed the importance of the standardization of variables
to proceed with the linear and other multivariate analysis.
Except for M_HDI and VHCLS, which presented a positive linear coefficient
of correlation equal to 0.75, all the other coefficients were below 0.50. But some
negative coefficients of correlation such as both morbidity rates with DD, GDP,
M_HDI, SNT and URB were noticed. This might prove a return on investments in
sanitation and urbanization. However, no significant linear correlation was observed
between arborization and other variables, except with DD. This means there is a
decrease in the number of urban households in streets with trees, as urban occupation
density increases. RD and DD showed to be positively correlated. Also, the cluster
analysis, both hierarchical (logic tree) and non-hierarchical (k-means), grouped both
rates of morbidities in the same cluster, and the other variables in another cluster. It
indicates that morbidities might be explained by such variables. Regarding the main
component analysis, it was found that four factors explain 71.78% of the variance
of the original (standardized) variables. The green infrastructure indicator (ARB)
showed the best load in factor 2, aligned with RD and CD.
Although results of multivariate analysis were encouraging, the classification
model, however, did not reveal rules with high accuracy and support. The analysis should be deeper, considering climatic variables, seasonality, sub populations,
different ranges of ages and genders, or excluding highly correlated variables, as
well as other hotspots to study urban management issues such as different mesoregions and metropolitan areas. Other modeling attempts might also involve extracting
rules of the main components instead of the original variables.
507
4 Conclusions: Classification Models Support Planning
and Public Policies for Urban Environment Towards
Sustainability
This study demonstrated the potential of data mining techniques to extract information of fairly easily available municipal indicators, such as demographic density (DD), Gross Domestic Product (GDP), Municipal Human Development Index
(M_HDI), percentages of households in streets with adequate sanitation (SNT), with
trees (ARB) and with urbanization (URB), and also the number of vehicles (VHCLS)
and morbidities of respiratory (RD) and circulatory (CD) diseases, responsible for
high rates of causes of death and hospitalizations in the State of Paraná and in Brazil.
This can help establish better urban management practices and reinforce some public
policies.
The variable that showed the largest variability was DD and the one with the
least variability was M_HDI. Most of the municipalities in the State of Paraná are
classified as small, with up to 50,000 inhabitants, and 70% (about 280 of 399) of the
municipalities presented DD equal to 3 to about 35 inhabitants per square kilometer. Meanwhile, the density of the largest cities in Paraná—Londrina, Maringá and
Curitiba (main city)—was 306.52, 733.14 and 4027.04 inhabitants per square kilometer, respectively; whereas Rio de Janeiro and São Paulo, the largest in the country,
counted 5265.82 and 7398.26 inhabitants per square kilometer, respectively, in the
same census of 2010. This revealed the importance of the standardization of variables
to proceed with the linear and other multivariate analysis.
Except for M_HDI and VHCLS, which presented a positive linear coefficient
of correlation equal to 0.75, all the other coefficients were below 0.50. But some
negative coefficients of correlation such as both morbidity rates with DD, GDP,
M_HDI, SNT and URB were noticed. This might prove a return on investments in
sanitation and urbanization. However, no significant linear correlation was observed
between arborization and other variables, except with DD. This means there is a
decrease in the number of urban households in streets with trees, as urban occupation
density increases. RD and DD showed to be positively correlated. Also, the cluster
analysis, both hierarchical (logic tree) and non-hierarchical (k-means), grouped both
rates of morbidities in the same cluster, and the other variables in another cluster. It
indicates that morbidities might be explained by such variables. Regarding the main
component analysis, it was found that four factors explain 71.78% of the variance
of the original (standardized) variables. The green infrastructure indicator (ARB)
showed the best load in factor 2, aligned with RD and CD.
Although results of multivariate analysis were encouraging, the classification
model, however, did not reveal rules with high accuracy and support. The analysis should be deeper, considering climatic variables, seasonality, sub populations,
different ranges of ages and genders, or excluding highly correlated variables, as
well as other hotspots to study urban management issues such as different mesoregions and metropolitan areas. Other modeling attempts might also involve extracting
rules of the main components instead of the original variables.