HAMIDA, ALYA (2025) ANALISIS PERBANDINGAN ALGORITMA MACHINE LEARNING DAN DEEP LEARNING UNTUK PERAMALAN PENJUALAN (STUDI KASUS: UMKM MINUMAN). S1 thesis, Universitas Mercu Buana.
|
Text (HAL COVER)
01 COVER.pdf Download (447kB) | Preview |
|
![[img]](https://repository.mercubuana.ac.id/style/images/fileicons/text.png) |
Text (BAB I)
02 BAB 1.pdf Restricted to Registered users only Download (209kB) |
|
|
Text (BAB II)
03 BAB 2.pdf Restricted to Registered users only Download (456kB) |
|
|
Text (BAB III)
04 BAB 3.pdf Restricted to Registered users only Download (475kB) |
|
|
Text (BAB IV)
05 BAB 4.pdf Restricted to Registered users only Download (819kB) |
|
|
Text (BAB V)
06 BAB 5.pdf Restricted to Registered users only Download (200kB) |
|
|
Text (DAFTAR PUSTAKA)
07 DAFTAR PUSTAKA.pdf Restricted to Registered users only Download (182kB) |
|
|
Text (LAMPIRAN)
08 LAMPIRAN.pdf Restricted to Registered users only Download (933kB) |
Abstract
Sales forecasting is a crucial component in the operations of small and medium enterprises (SMEs), especially in the beverage industry, which is highly susceptible to fluctuating consumer demand. This study aims to compare the performance of Machine Learning algorithms (Random Forest and LightGBM) with Deep Learning algorithms (Long Short-Term Memory/LSTM and Temporal Convolutional Network/TCN) in predicting daily sales at two outlets of a beverage SME. The research adopts a quantitative approach with time series forecasting methods. Daily sales data were collected over a two-year period (2023–2024), resulting in over 20,000 transaction records. The preprocessing stage included data normalization, addition of contextual features (weather and promotional days), and evaluation under three data split scenarios (80:20, 70:30, and 60:40). Model performance was evaluated using Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). The findings reveal that LSTM performed best on the Bakti outlet (MAE: 3.12, RMSE: 3.83) under the 80:20 split scenario, while LightGBM yielded optimal results on the Tugu outlet (MAE: 6.14, RMSE: 19.87). Deep learning models demonstrated higher sensitivity to training data volume, whereas machine learning models offered greater stability but lower peak performance. These insights highlight the importance of selecting forecasting approaches that align with the data characteristics of each outlet. Keywords: sales forecasting, machine learning, deep learning, LSTM, LightGBM, SMEs, time series. Peramalan penjualan merupakan elemen krusial dalam operasional UMKM, khususnya dalam industri minuman yang memiliki fluktuasi permintaan tinggi. Penelitian ini bertujuan untuk membandingkan performa algoritma Machine Learning (Random Forest dan LightGBM) dengan algoritma Deep Learning (LSTM dan Temporal Convolutional Network/TCN) dalam memprediksi penjualan harian pada dua outlet Kedai Minuman. Data penjualan harian dikumpulkan selama dua tahun (2023–2024), dengan total lebih dari 20.000 transaksi. Penelitian ini menggunakan pendekatan kuantitatif dengan metode time-series forecasting. Proses preprocessing mencakup normalisasi data, penambahan fitur kontekstual (cuaca dan hari promosi), serta pengujian pada tiga skenario pembagian data (80:20, 70:30, dan 60:40). Evaluasi performa model dilakukan menggunakan metrik Mean Absolute Error (MAE) dan Root Mean Squared Error (RMSE). Hasil penelitian menunjukkan bahwa model LSTM menghasilkan performa terbaik pada outlet Bakti (MAE: 3.12, RMSE: 3.83) pada skenario 80:20, sementara LightGBM menunjukkan hasil optimal di outlet Tugu (MAE: 6.14, RMSE: 19.87). Model deep learning menunjukkan sensitivitas terhadap volume data latih, sementara model machine learning cenderung lebih stabil namun kurang unggul dalam skenario optimal. Temuan ini memberikan implikasi bagi UMKM untuk memilih pendekatan prediktif yang sesuai dengan karakteristik data mereka. Kata Kunci: peramalan penjualan, machine learning, deep learning, LSTM, LightGBM, UMKM, time series.
Actions (login required)
 |
View Item |
