Pengembangan Algoritma Peramalan Untuk Aplikasi Di teknik Industri
ABSTRACT This paper aims to present partial results of a research on developing an algorithm to select the most suitable forecasting techniques applied in industrial engineering applications. This paper emphasizes on the assessment of various time-series forecasting techniques which are applied to r...
Saved in:
| 主要作者: | |
|---|---|
| 格式: | Article NonPeerReviewed |
| 出版: |
[Yogyakarta] : Universitas Gadjah Mada
2005
|
| 主題: | |
| 在線閱讀: | https://repository.ugm.ac.id/23368/ http://i-lib.ugm.ac.id/jurnal/download.php?dataId=6315 |
| 標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
| 機構: | Universitas Gadjah Mada |
| 總結: | ABSTRACT
This paper aims to present partial results of a research on developing an algorithm to select the most suitable forecasting techniques applied in industrial engineering applications. This paper emphasizes on the assessment of various time-series forecasting techniques which are applied to real-world time-series data.
To assess the forecasting techniques, the first step was collecting time-series data from various sources, including Badan Pusat Statistik (Indonesian Bureau of Statistics) and manufacturing industries. The second step was selecting readily available time-series forecasting techniques. They include moving average, exponential smoothing (Halts-Winters), classic decomposition, Fourier series, ARIMA (Box-Jenkins) and linear trend. Having selected the techniques, the models of forecast for each time-series data were developed. The models were then validated and analyzed using some forecasting accuracy measures namely: MAD, MSE, MAPE and MPE. The forecast results were also analyzed using tracking signal control map.
The research found some interesting findings. Firstly, forecasting accuracy measures can not be used as a single reference on determining the most suitable forecasting technique. Tracking signal must also be applied in conjunction with the accuracy measures. It was also found that time-series data (with random pattern, without trend element) was best forecasted by using exponential smoothing. The value of a will be nearly one for data which has cyclical and seasonality elements dominated the pattern. The value of will be close to zero if there was no dominant pattern observed Box-Jenkins method was preferred to forecasting Index Harga Saham Gabungcm (IHSG) data.
Keywords: forecasting, time-series, industrial engineering applications, algorithm |
|---|