| Summary: | Survey Base of National Education year 2003 (SDPN 2003) is a data
realization of education mapping in Indonesia; it gave a large education
database with 3,89 GB (4,178,499.369 bytes) and 2,395 files involved 203,590
records and 569 indicators, also it has a measured variables, high dimension with
a hundreds heterogeneous attribute and distributed spatial data geographically .
Data mining can be used to extract knowledge from a large database. Data
mining develop to be spatial data mining to extract knowledge from spatial
relation to get some of explicit pattern which is not found in database. Because of
Indonesia has a wide location distribution: provinces, cities, and districts which
have many social culture, so data mining can be applied in mapping of quality
education in Indonesia.
The aims of this research are to study and to apply spatial data mining for
classification the quality of education at various rates in Indonesia using:
Expansion Spatial AutoRegressive-Kriging (SAR-Kriging) which is combine
expansion SAR model and Kriging method. The development of SAR-Kriging
model theoretically based on the weaknesses of SAR model, it is only applicable
for prediction at sample locations. In the other side Kriging method can be used
to predict observations at unsample locations. The expansion SAR model is an
extension of model SAR to give information the influence of total independent
variables to dependent variable through spatial heterogeneity using locations
coordinate. The combination of expansion SAR model and Kriging method is
studied to get the causal model which can be used to predict at unsample
locations. Using SAR-Kriging model we can predict the quality of education at
unsample locations in Indonesia at province or district level.
Methodology in this research is a data mining processes and knowledge
discovery in database (KDD) using three stages. The first stage is preprocessing
data involve database preparation of Survey Base of National Education year
2003 (SDPN 2003). SDPN 2003 is a realization of education mapping in
Indonesia. In SDPN 2003 we have a complex and large data, especially for
education at elementary, junior and senior schools. We have a large school
database 203,590 records and 569 indicators. It can be divided for elementary
school (SD) 158,590 records and 122 indicators, junior high school (SMP) 28,949
records and 138 indicators, for senior high school (SMA) 10,810 records and 142
indicators, and SMK 5,156 records and 167 indicators. Furthermore, in the first
steps the data is cleaned and transformed, variable is selected by factor analysis
and Structural Equation Model (SEM), and the last is combining spatial and non
spatial data. The second stage is data mining, using Moran index, SAR,
Expansion SAR and SAR-Kriging models to get description and prediction of
quality education. The last stage is post processing through interpretation,
evaluation and visualization to get knowledge.
The result of data processing of SDPN 2003 using SAR-Kriging model shows
that the prediction of quality education of elementary school (SD), junior high
xxiii
school (SMP) and senior high school (SMA) at twelve districts at West Java
Province has Mean Absolute Percentage Error (MAPE) 8.39; 23.63 and 40.45. It
shows that the SAR-Kriging model has MAPE less than 10% for elementary
school, so it fit forprediction of education quality at SD. The processing data at
provinces in Java Island gives the MAPE of SD, SMP and SMA: 8.15; 8.10 and
29.36. The result shows that the SAR-Kriging model can be used to predict the
quality of education for SD and SMP at unsample provinces in Java Island.
Using SAR-Kriging we also predict the quality of education at province Aceh
which gives MAPE for SD, SMP and SMA. It gives the MAPE 8.44; 14.51 and
42.25. It shows that the SAR-Kriging model can be used to predict quality of
education SD at Aceh province. For the extension province as West Sulawesi, we
get the MAPE of SAR-Kriging model for SD, SMP and SMA are 1.02; 2.19 and
81.76. So, at West Sulawesi the SAR-Kriging model also fit to the data SD and
SMP. The processing data for 13 provinces in Indonesia for SD, SMP and SMA
give SAR-Kriging MAPE 8.07; 7.48 and 29.21. We can conclude that the SARKriging model is a good model for prediction of quality of education for SD and
SMP. All of the results show that SAR-Kriging model at elementary and junior
high schools is a good model for prediction a quality of education at unsample
provinces in Indonesia.
To get processing data easier we built the application software of Spatial Data
Mining using SAR, Expansion SAR and SAR-Kriging models. Using this software
we can apply the model to get description and prediction of quality education at
various locations, districts or provinces in Indonesia. The result of prediction of
quality education using SAR-Kriging can be used as a recommendation for
management of education in Indonesia.
|
|---|