RAMBING, Danni Hastanto and Kusumaningrum, Retno and Sugiharto, Aris (2025) OPTIMISASI PERAMALAN PENJUALAN DALAM UKM RITEL: PENDEKATAN HIERARCHICAL RECONCILIATION DEEP LEARNING UNTUK MANAJEMEN PERSEDIAAN YANG EFEKTIF. Masters thesis, UNIVERSITAS DIPONEGORO.

	Text Cover aja.pdf Download (159kB)
	Text COVER FIX.pdf Restricted to Repository staff only Download (5MB)
	Text BAB I.pdf Download (317kB)
	Text BAB II.pdf Download (865kB)
	Text BAB III.pdf Restricted to Repository staff only Download (741kB)
	Text BAB IV.pdf Restricted to Repository staff only Download (3MB)
	Text BAB V.pdf Restricted to Repository staff only Download (291kB)
	Text Daftar Pustaka.pdf Download (292kB)
	Text Lampiran.pdf Restricted to Repository staff only Download (692kB)

Abstract

Usaha Kecil dan Menengah (UKM) memiliki peran strategis dalam perekonomian namun menghadapi tangan yang signifikan dalam manajemen persediaan produk karena semakin banyaknya item dalam sistem inventaris ritel, membuat peramalan akurat menjadi kompleks. Beragam teknik peramalan telah dikembangkan, baik metode statistik klasik, machine learning, hingga deep learning. Akan tetapi, pendekatan ini umumnya dirancang untuk melakukan peramalan pada satu tingkat hierarki tertentu, seperti total penjualan, kategori produk, atau tingkat SKU (Stock Keeping Unit). Strategi peramalan pendekatan Top-Down dan Bottom-Up tidak dapat memanfaatkan keterkaitan antar runtun waktu hierarki. Penelitian ini mengusulkan model deep learning NBEATS (Neural Basis Expansion Analysis for interpretable Time Series forecasting) dipadukan dengan rekonsiliasi hierarki MinT non-negatif. Tujuan dari penelitian ini adalah mengevaluasi akurasi model dasar NBEATS dibandingkan dengan model dasar ETS dan ARIMA; mengevaluasi akurasi penerapan metode rekonsiliasi hierarkis terhadap masing‑masing model dasar; dan menentukan kombinasi model dasar dan rekonsiliasi paling optimal pada data penjualan harian Funan Mart. Hasil penelitian menunjukkan model dasar NBEATS SMAPE 21,88%, lebih unggul dibanding ETS (41,09%) dan ARIMA (38,26%). Penerapan rekonsiliasi menurunkan SMAPE masing‑masing model dasar, 13,995% (ETS), 0,599% (ARIMA), dan 2,006% (NBEATS). Kombinasi NBEATS dengan rekonsiliasi WLS non-negatif menghasilkan peramalan lebih akurat dengan nilai SMAPE 21,441%. Temuan ini mengkonfirmasi efektivitas pendekatan yang diusulkan dalam meningkatkan akurasi peramalan produk pada UKM ritel.
Kata Kunci : UKM Ritel, Peramalan Runtun Waktu, NBEATS, ETS, ARIMA, Rekonsiliasi Hierarkis

Small and medium-sized enterprises (SMEs) play a strategic role in the economy but face significant challenges in product inventory management due to the increasing number of items in the retail inventory system, making accurate forecasting complex. Various forecasting techniques have been developed, including classical statistical methods, machine learning, and deep learning. However, these approaches are generally designed to forecast at a certain hierarchical level, such as total sales, product categories, or product SKU (Stock Keeping Unit) levels. Top-down and Bottom-Up forecasting strategies cannot utilize the relationship between hierarchical time series. This study proposes a deep learning NBEATS (Neural Basis Expansion Analysis for interpretable Time Series forecasting) model combined with non-negative MinT hierarchical reconciliation. The objectives of this study are to evaluate the accuracy of the NBEATS baseline model compared to the ETS and ARIMA baseline models, evaluate the accuracy of the application of the hierarchical reconciliation method to each baseline model, and determine the most optimal combination of baseline and reconciliation models on Funan Mart daily sales data. The results show that the NBEATS baseline model has a SMAPE of 21.88%, superior to ETS (41.09%) and ARIMA (38.26%). The application of reconciliation decreases the SMAPE of each base model by 13.995% (ETS), 0.599% (ARIMA), and 2.006% (NBEATS). The combination of NBEATS with non-negative WLS reconciliation produces more accurate forecasting with an SMAPE value of 21.441%. This finding confirms the effectiveness of the proposed approach in improving product forecasting accuracy in retail SMEs.
Keyword : Retail SMEs, Time Series Forecasting, NBEATS, ETS, ARIMA, Hierarchical Reconciliation

Actions (login required)

View Item