PREDIKSI PENJUALAN UMKM MENGGUNAKAN ALGORITMA XGBOOST UNTUK MENDUKUNG PENGAMBILAN KEPUTUSAN BISNIS PADA DATA TRANSAKSI PERCETAKAN DIGITAL
Digital transaction records can help micro, small, and medium enterprises (MSMEs) forecast sales, but their usefulness depends on forecast errors and the ability to anticipate sales spikes. This study evaluates XGBoost for one-day-ahead sales value forecasting at XYZ Print. A total of 8,808 line items from 5,547 invoices dated from 2 May 2024 to 31 October 2025 were aggregated into 548 calendar dates, of which 526 had transaction records. The model used calendar features, historical sales values, and historical invoice counts. Ninety-six configurations were screened through five temporal validation folds, and the selected model was evaluated on 90 days with observed sales from August to October 2025. XGBoost achieved a mean absolute error (MAE) of IDR 1,208,167, a root mean squared error (RMSE) of IDR 2,488,751, and a weighted absolute percentage error (WAPE) of 53.85%. Its MAE was 2.04% below that of the same-weekday median, the strongest benchmark. The five highest-sales days accounted for 81.35% of the model’s total squared error. The forecasts can serve as an initial reference for routine sales planning, while large orders warrant a separate review using confirmed bookings. Keywords: sales forecasting, XGBoost, MSMEs, digital printing, decision-making Digital transaction records can help micro, small, and medium enterprises (MSMEs) forecast sales, but their usefulness depends on forecast errors and the ability to anticipate sales spikes. This study evaluates XGBoost for one-day-ahead sales value forecasting at XYZ Print. A total of 8,808 line items from 5,547 invoices dated from 2 May 2024 to 31 October 2025 were aggregated into 548 calendar dates, of which 526 had transaction records. The model used calendar features, historical sales values, and historical invoice counts. Ninety-six configurations were screened through five temporal validation folds, and the selected model was evaluated on 90 days with observed sales from August to October 2025. XGBoost achieved a mean absolute error (MAE) of IDR 1,208,167, a root mean squared error (RMSE) of IDR 2,488,751, and a weighted absolute percentage error (WAPE) of 53.85%. Its MAE was 2.04% below that of the same-weekday median, the strongest benchmark. The five highest-sales days accounted for 81.35% of the model’s total squared error. The forecasts can serve as an initial reference for routine sales planning, while large orders warrant a separate review using confirmed bookings. Keywords: sales forecasting, XGBoost, MSMEs, digital printing, decision-making
Authors
- Azzahra Najwa Ali Rambe
Publication Details
- Journal
- E-Mabis/E-Mabis: Jurnal Ekonomi Manajemen dan Bisnis
- Published
- 2026-11-21
- DOI
- https://doi.org/10.29103/e-mabis.v27i2.1647
- Primary Topic
- Forecasting Techniques and Applications
- Type
- article
- Field-Weighted Citation Impact
- 0.00