Prediksi Penjualan Harian Tiktokshop Menggunakan Hybrid Stacking Regressor: Studi Kasus OnefreshID

Jauhari, Mohammad Yusron (2026) Prediksi Penjualan Harian Tiktokshop Menggunakan Hybrid Stacking Regressor: Studi Kasus OnefreshID. Masters thesis, Institut Teknologi Sepuluh November.

Text 6032242104-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (1MB) | Request a copy

Abstract

Prediksi penjualan harian di TikTok Shop menghadapi tantangan akibat fluktuasi interaksi pengguna dan variasi pola temporal yang memengaruhi perilaku belanja. Keterbatasan data historis, ketidakseimbangan pola permintaan, serta dinamika faktor temporal berpotensi menyebabkan model peramalan kurang optimal dalam menangkap pola penjualan harian. Oleh karena itu, penelitian ini mengintegrasikan variabel interaksi pengguna yang meliputi jumlah suka, komentar, bagikan, dan penayangan dengan fitur temporal untuk membangun model prediksi penjualan harian. Metode yang diusulkan adalah Hybrid Stacking Regressor, yang mengombinasikan beberapa algoritma regresi, yaitu LSTM, CNN 1D, dan Random Forest. Dataset yang digunakan berupa data penjualan harian OnefreshID selama satu tahun yang dipadankan dengan metrik interaksi pengguna dan fitur kalender. Evaluasi model dirancang melalui perbandingan antara model menggunakan variabel interaksi pengguna dan model yang mengintegrasikan variabel interaksi–temporal, dengan metrik evaluasi MAE dan RMSE. Selain itu, Permutation Importance dan SHAP digunakan untuk menganalisis kontribusi masing-masing fitur terhadap prediksi model. Penelitian ini berkontribusi dengan mengkaji integrasi variabel interaksi pengguna dan faktor temporal dalam kerangka Hybrid Stacking Regressor pada konteks TikTok Shop Indonesia, yang masih terbatas dibahas dalam penelitian terdahulu, serta diharapkan dapat menjadi dasar pengambilan keputusan terkait perencanaan stok dan penjadwalan konten berbasis data.
=========================================================================================================================================
Daily sales predictions on TikTok Shop face challenges due to fluctuations in user interactions and variations in temporal patterns that affect shopping behavior. Limited historical data, imbalances in demand patterns, and the dynamics of temporal factors have the potential to cause forecasting models to be less than optimal in capturing daily sales patterns. Therefore, this study integrates user interaction variables, including the number of likes, comments, shares, and views, with temporal features to build a daily sales prediction model. The proposed method is Hybrid Stacking Regressor, which combines several regression algorithms, namely LSTM, CNN 1D, and Random Forest. The dataset used consists of OnefreshID's daily sales data for one year, matched with user interaction metrics and calendar features. Model evaluation is designed by comparing models using user interaction variables and models that integrate interaction-temporal variables, with MAE and RMSE evaluation metrics. Additionally, Permutation Importance and SHAP were used to analyze the contribution of each feature to the model's prediction. This research contributes by examining the integration of user interaction variables and temporal factors within the Hybrid Stacking Regressor framework in the context of TikTok Shop Indonesia, which has been limited in previous studies, and is expected to serve as a basis for decision-making related to inventory planning and data-driven content scheduling.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	TikTok Shop, interaksi pengguna, faktor temporal, prediksi penjualan, Hybrid Stacking Regressor
Subjects:	H Social Sciences > HD Industries. Land use. Labor > HD30.27 Business forecasting
Divisions:	Interdisciplinary School of Management and Technology (SIMT) > 61101-Master of Technology Management (MMT)
Depositing User:	Mohammad Yusron Jauhari
Date Deposited:	29 Jul 2026 08:07
Last Modified:	29 Jul 2026 08:07
URI:	http://repository.its.ac.id/id/eprint/139727

Actions (login required)

View Item