THE HYBRID MODEL OF NEURAL NETWORK AND LINEAR REGRESSION FOR AGRICULTURAL COMMODITY PRICE FORECASTING IN INDONESIA
This study aims to develop a short-term forecasting model of agricultural commodity prices using the traditional-intelligence hybrid model. The main purpose of using the hybrid model was its ability to accommodate linear and non-linear patterns in commodity prices, thereby increasing the model'...
Saved in:
| Main Author: | |
|---|---|
| Format: | Theses |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/71455 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | This study aims to develop a short-term forecasting model of agricultural commodity prices using the traditional-intelligence hybrid model. The main purpose of using the hybrid model was its ability to accommodate linear and non-linear patterns in commodity prices, thereby increasing the model's accuracy. Four agricultural commodities were used as case studies: shallots, garlic, red chillies, and cayenne pepper.
Agricultural commodity prices were causally predicted using the keyword search index from Google Trends, macroeconomic parameters, and demand-supply parameters. Several steps were performed before modelling: pre-processing, root test with Dickey-Fuller, lag selection using the Auto Regressive Distributed Lag method, time-series variables addition, data normalization, and variable selection using the LASSO regression method. Modelling was then performed by separating commodity prices' linear and non-linear components. The linear component was modelled by linear regression and the non-linear component by the averaged neural network model, which has an input layer, a hidden layer, and an output layer.
Model performance was assessed using the MAPE value parameter. The result showed that the model had high accuracy with MAPE value below 10%. Application of the model also resulted in better performance compared to previous studies and time-series methods. On the other hand, the model had limitations in interpreting the importance of the predictor variables since there were linear and non-linear components, each having a different level of importance of the variable. The developed model also did not consider weather factors due to limited data, so it is hoped that it can be included in further research. Nevertheless, with high accuracy, the developed model could be an additional input for designing government policies and strategies for business people in agriculture.
|
|---|