A new hybrid genetic algorithm-sarima-artificial neural network in forecasting Malaysian export amount of palm oil
Malaysia is a significant export country of palm oil to all over the world. Therefore, forecasting of palm oil export is required to help in boosting the nation’s socioeconomic development as well as for the plantation companies to sustain and improve for a better management regarding export. T...
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English English |
| Published: |
2021
|
| Subjects: | |
| Online Access: | http://eprints.uthm.edu.my/6292/1/24p%20CHAI%20KAH%20CHUN.pdf http://eprints.uthm.edu.my/6292/2/CHAI%20KAH%20CHUN%20COPYRIGHT%20DECLARATION.pdf http://eprints.uthm.edu.my/6292/3/CHAI%20KAH%20CHUN%20WATERMARK.pdf |
| Summary: | Malaysia is a significant export country of palm oil to all over the world. Therefore,
forecasting of palm oil export is required to help in boosting the nation’s socioeconomic
development as well as for the plantation companies to sustain and
improve for a better management regarding export. The traditional Seasonal
Autoregressive Integrated Moving Average (SARIMA) model assumes that all the
parameters in the non-seasonal and seasonal parameters are significant which will
lead to inaccuracy in the model identification stage and increase the cost of reidentification.
Hence, this study aimed to optimise the order of subset of the
SARIMA model using Genetic Algorithm (GA). It would be then combined with
Artificial Neural Network (ANN) to form a novel hybrid GA-SARIMA-ANN to
predict Malaysian palm oil export. The performance of the proposed hybrid GASARIMA-ANN
was
compared
with
four
existing
models,
which
were
SARIMA,
ANN,
hybrid
GA-SARIMA,
and
hybrid
SARIMA-ANN.
The
forecast
accuracy
for
all
the models was evaluated using Mean Absolute Error (MAE), Root Mean
Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Pearson
correlation coefficient. The empirical result showed that the proposed hybrid GASARIMA-ANN
(5-6-1)
yielded
the
lowest
forecasting
accuracy
where
its
MAPE
was
only 8.15% compared with the existing models, whose MAPE values were
slightly above 10%. In addition, the proposed hybrid model achieved the highest
correlation coefficient, higher by 27% on average compared with the benchmark
models. It is proved that GA facilitates the model identification for SARIMA, while
the coupling of ANN may help to model nonlinearity, making the hybrid model’s
forecast more accurate. |
|---|