Please use this identifier to cite or link to this item: http://repository.ipb.ac.id/handle/123456789/171520
Title: Time Series Clustering dengan Mahalanobis Distance-Based Dynamic Time Warping pada Peramalan Tingkat Gerombol Harga Bawang Merah Provinsi di Indonesia
Other Titles: 
Authors: Fitrianto, Anwar
Wigena, Aji Hamim
Kereh, Pingkan Febbe Fiorela
Issue Date: 2025
Publisher: IPB University
Abstract: Bawang merah menjadi salah satu komoditas hortikultura strategis yang memiliki tingkat permintaan yang cukup tinggi di Indonesia dan produksinya dinilai sangat lambat karena sifatnya yang musiman. Hal tersebut berdampak pada terjadinya fluktuasi harga yang menyebabkan harga bawang merah menjadi tidak stabil. Penelitian ini bertujuan menggerombolkan provinsi berdasarkan pola pergerakan harga bawang merah yang serupa melalui pendekatan time series clustering dengan jarak Mahalanobis distance-based dynamic time warping (MDDTW) dan melakukan peramalan tingkat gerombol menggunakan metode seasonal autoregressive integrated moving average (SARIMA). Data yang digunakan dalam penelitian ini adalah harga bawang merah mingguan dari 34 provinsi di Indonesia yang bersumber dari Portal Satu Data Kementerian Perdagangan Republik Indonesia. Analisis data dilakukan dengan bantuan software Rstudio. Hasil penelitian menunjukkan pautan rataan sebagai pautan terbaik yang menggerombolkan data harga bawang merah dari 34 provinsi ke dalam tiga gerombol. Pemodelan deret waktu dilakukan pada data prototype setiap gerombol. Dari keseluruhan model tentatif setiap prototype, dipilih dua kandidat model terbaik berdasarkan nilai AIC terkecil. Seluruh model tersebut dievaluasi performanya dengan menerapkan skema expanding window dan diperoleh nilai rataan MAPE dari setiap model. Hasil peramalan setiap gerombol menunjukkan akurasi yang baik, dengan nilai rataan MAPE di bawah 20%.
Shallot are one of the strategic horticultural commodities that have a fairly high demand in Indonesia, and their production is considered very slow due to their seasonal nature. This has an impact on price fluctuations, causing shallot prices to become unstable. This study aims to cluster provinces based on similar shallot prices movement patterns using a time series clustering approach with Mahalanobis distance-based dynamic time warping (MDDTW) and to forecast cluster level using the seasonal autoregressive integrated moving average (SARIMA) method. The data used in this study are weekly shallot prices from 34 provinces in Indonesia sourced from the One Data Portal The Ministry of Trade Republic of Indonesia. Data analysis was performed using Rstudio software. The results show that the average linkage is the best linkage that clusters shallot price data from 34 provinces into three clusters. Time series modeling was performed on the prototype data of each cluster. From all tentative models of each prototype, the two best model candidates were selected based on the smallest AIC value. All models were evaluated for performance by applying an expanding window scheme, and the average MAPE value of each model was obtained. The forecasting results for each cluster showed good accuracy, with an average MAPE value below 20%.
URI: http://repository.ipb.ac.id/handle/123456789/171520
Appears in Collections:UT - Statistics and Data Sciences

Files in This Item:
File Description SizeFormat 
cover_G1401211087_b60fc8e1728b4a8bb4984772260fad2f.pdfCover601.27 kBAdobe PDFView/Open
fulltext_G1401211087_5e6e372b3e6945a59d3a5d456d49e1fa.pdf
  Restricted Access
Fulltext1.57 MBAdobe PDFView/Open
lampiran_G1401211087_593daac398784d4f80a305faf5dd4735.pdf
  Restricted Access
Lampiran961.33 kBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.