| dc.description.abstract | CPI is one of the most frequently used
indicators to measure the inflation rate in a region. The
government can maintain economic stability by
knowing the CPI value in advance. Therefore, we need
a suitable method to predict an accurate CPI value. In
this research, we investigate the prediction of CPI
based on the machine learning method, SVR, and
compare it to the ARIMAX method. We use Indonesia
CPI data from January 2015 to October 2021. We
investigate the SVR method using four kernel
functions: Radial Basis Function (RBF), Polynomial,
Linear, and Sigmoid. We build the ARIMAX model
through the auto ARIMA process. We divide the data
into two parts with three scenarios to investigate the
performance of the methods: training and testing. The
results show that the partition of 80% training and
20% testing gives the best performance. The SVR
method performs best using a linear kernel, with an
RMSE value of 0,743 and a MAPE value of 0,684%.
The best ARIMAX model is model (0,2,1) with an
RMSE value of 1,928071 and a MAPE of 1,731598 %.
From the plot of prediction results and indicators of
RMSE and MAPE, the SVR predicts CPI data bettethan the ARIMAX method, with CPI in the previous
one-month data (MA1) being the most influential
variable on the next CPI value | en_US |
