Pengembangan Model Prediksi Permintaan Baby Product Pada Merchant Shopee Menggunakan Support Vector Regression

Devi, Echa Karlinda (2026) Pengembangan Model Prediksi Permintaan Baby Product Pada Merchant Shopee Menggunakan Support Vector Regression. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6032241204-Master_ Thesis.pdf Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Pasar popok dan susu bayi di Indonesia menunjukkan tren pertumbuhan yang signifikan dengan proyeksi nilai masing-masing mencapai USD 2,39 miliar dan USD 5,35 miliar pada tahun 2029. Namun, pelaku bisnis e-commerce, seperti salah satu merchant di Shopee yang menjadi objek penelitian ini, menghadapi tantangan berupa fluktuasi permintaan yang sangat tajam dan tidak menentu. Data menunjukkan lonjakan penjualan dari Rp 30,1 juta pada kuartal 1 tahun 2023 menjadi Rp 576 juta pada kuartal 1 tahun 2024, sebelum akhirnya menurun drastis menjadi Rp 265 juta pada kuartal 4 tahun 2024. Ketidakpastian ini dipicu oleh variabel dinamis seperti event promosi dan tren pasar yang tidak mampu ditangani secara optimal oleh metode statistik tradisional seperti ARIMA. Penelitian ini bertujuan untuk membangun model prediksi permintaan yang akurat menggunakan metode Support Vector Regression (SVR) guna mengantisipasi lonjakan atau penurunan permintaan serta memberikan rekomendasi strategi manajemen inventori dan pemasaran yang efektif. Variabel yang digunakan mencakup fitur transaksional (SKU, kategori, harga, berbagai jenis potongan harga/voucher, koin Shopee, dan diskon ongkir) serta faktor eksternal (hari libur nasional dan event Shopee). Data diproses melalui tahap cleaning, feature engineering, dan normalisasi, kemudian dilakukan uji perbandingan dimana uji coba pertama menggunakan data latih (70%) dan data uji (30%), lalu dilakukan uji coba kedua dengan menggunakan data latih (80%) dan data uji (20%). Optimasi model dilakukan menggunakan kernel RBF dengan teknik Grid Search untuk menemukan hyperparameter paling optimal. Hasil penelitian menunjukkan bahwa model pada uji coba kedua menggunakan perbandingan data latih (80%) dan data uji (20%) menghasilkan hasil yang lebih baik dimana mampu menjelaskan sekitar 52% variabilitas data permintaan dengan nilai MAE sebesar 1,34 unit dan RMSE sebesar 3,73 unit. Kesimpulannya, model ini cukup andal dalam menangkap pola umum dan tren fluktuasi permintaan di pasar yang volatil. Implementasi model dalam bentuk dashboard interaktif memberikan manfaat praktis bagi merchant untuk melakukan simulasi skenario promosi dan mengoptimalkan pengelolaan stok, sehingga dapat meningkatkan efisiensi operasional dan profitabilitas bisnis jangka panjang.
========================================================================================================================
The diaper and infant formula markets in Indonesia show a significant growth trend, with projected values reaching USD 2.39 billion and USD 5.35 billion respectively by 2029. However, e-commerce business players—such as a Shopee merchant that serves as the object of this study—face challenges in the form of highly volatile and unpredictable demand fluctuations. The data indicate a surge in sales from IDR 30.1 million in the first quarter of 2023 to IDR 576 million in the first quarter of 2024, followed by a sharp decline to IDR 265 million in the fourth quarter of 2024. This uncertainty is driven by dynamic variables such as promotional events and market trends, which cannot be optimally handled by traditional statistical methods such as ARIMA. This study aims to develop an accurate demand forecasting model using the Support Vector Regression (SVR) method to anticipate spikes or declines in demand and to provide effective inventory management and marketing strategy recommendations. The variables include transactional features (SKU, category, price, various types of discounts/vouchers, Shopee coins, and shipping discounts) as well as external factors (national holidays and Shopee events). The data are processed through data cleaning, feature engineering, and normalization stages, followed by comparative testing in two experiments: the first using a 70% training set and 30% test set, and the second using an 80% training set and 20% test set. Model optimization is conducted using an RBF kernel with the Grid Search technique to identify the most optimal hyperparameters. The results show that the second experiment, using an 80% training and 20% testing data split, yields better performance, explaining approximately 52% of the variability in demand data, with an MAE of 1.34 units and an RMSE of 3.73 units. In conclusion, the model is sufficiently reliable in capturing general patterns and demand fluctuation trends in a volatile market. The implementation of the model in the form of an interactive dashboard provides practical benefits for merchants by enabling promotional scenario simulations and inventory optimization, thereby improving operational efficiency and long-term business profitability.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Grid Search, Machine Learning, Perencanaan Stok, Prediksi Permintaan, Support Vector Regression (SVR), Demand Forecasting, Grid Search, Inventory Planning, Machine Learning, Support Vector Regression (SVR)
Subjects:	T Technology > T Technology (General) > T174 Technological forecasting
Divisions:	Interdisciplinary School of Management and Technology (SIMT) > 61101-Master of Technology Management (MMT)
Depositing User:	Devi Echa Karlinda
Date Deposited:	10 Feb 2026 06:59
Last Modified:	10 Feb 2026 06:59
URI:	http://repository.its.ac.id/id/eprint/132313

Actions (login required)

View Item