Please use this identifier to cite or link to this item: http://repository.ipb.ac.id/handle/123456789/171467
Title: Perbandingan Performa Model Lstm Multivarat Dengan Dan Tanpa Kovariat Dalam Peramalan Penjualan Minuman Kekinian
Other Titles: 
Authors: Indahwati
Anisa, Rahma
Tirtasuwanda, Aisyah Nuruzzahra
Issue Date: 2025
Publisher: IPB University
Abstract: Peramalan penjualan dalam industri makanan dan minuman berperan penting untuk mendukung pengambilan keputusan operasional, terutama pada bisnis dengan permintaan yang fluktuatif seperti kedai kopi. Penelitian ini menggunakan data penjualan harian enam menu minuman kekinian di Et Al Coffee Bogor selama periode 1 September 2021 hingga 30 Mei 2025, dengan total 8.208 observasi. Model Long Short-Term Memory (LSTM) multivariat diterapkan karena mampu menangkap pola temporal jangka panjang serta memproses lebih dari satu peubah penjelas secara bersamaan. Penelitian ini bertujuan untuk membandingkan performa model peramalan dengan dan tanpa kovariat sekaligus meramalkan penjualan selama 21 hari ke depan. Evaluasi dilakukan dengan walk-forward cross-validation, dengan kovariat yang meliputi indikator Ramadan, hari libur nasional, periode promosi, dan hari dalam minggu dalam bentuk cyclical encoding. Hasil menunjukkan bahwa penambahan kovariat meningkatkan performa model, dengan model terbaik mencapai nilai SMdAPE sebesar 38%. Peramalan 21 hari ke depan memperlihatkan pola penjualan yang konsisten dengan tren meningkat dari weekdays menuju weekend. Temuan ini memberikan pendekatan prediksi yang efisien dalam mengelola persediaan dan perencanaan strategi penjualan.
Sales forecasting in the food and beverage industry played an important role in supporting operational decision-making, especially for businesses with fluctuating demand such as coffee shops. This study used daily sales data of six popular beverages items from Et Al Coffee Bogor for the period between September 1, 2021, and May 30, 2025, comprising a total of 8,208 observations. A multivariate Long Short-Term Memory (LSTM) model was applied, as it was capable of capturing long-term temporal patterns and processing multiple explanatory variables simultaneously. The study aimed to compare the performance of forecasting models with and without covariates while forecasting sales for the next 21 days. Evaluation was conducted using walk-forward cross-validation, with covariates including Ramadan indicators, national holidays, promotion periods, and day-of-week variables represented through cyclical encoding. The results showed that incorporating covariates improved model performance, with the best model achieving an SMdAPE value of 38%. The 21-day-ahead forecast revealed consistent sales patterns with an increasing trend from weekdays to weekends. These findings provided an efficient predictive approach for inventory management and sales strategy planning.
URI: http://repository.ipb.ac.id/handle/123456789/171467
Appears in Collections:UT - Statistics and Data Sciences

Files in This Item:
File Description SizeFormat 
cover_G1401211076_c61a4048e3c74f08aadace5904ba88b3.pdfCover489.16 kBAdobe PDFView/Open
fulltext_G1401211076_1514962262e6490ea73c58618073d1d7.pdf
  Restricted Access
Fulltext1.6 MBAdobe PDFView/Open
lampiran_G1401211076_8de7df96aacc47ac9bf893efe43b2101.pdf
  Restricted Access
Lampiran2.17 MBAdobe PDFView/Open


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