Accurate demand forecasting is a critical enabler of digital transformation for fashion-retail Small and Medium Enterprises (SMEs), whose sales patterns are strongly shaped by seasonality, promotions, and rapidly shifting trends. Prior hybrid Random Forest–Support Vector Machine (RF-SVM) models have reported near-perfect accuracy in predicting retail transaction value, but were trained with default hyperparameters and were not evaluated on demand-volume forecasting. This study evaluates the extent to which systematic hyperparameter tuning improves the performance of an ensemble RF-SVM model when applied to weekly demand-volume prediction for fashion-retail SMEs. Weekly sales records from retail SMEs in the Surakarta region (Indonesia) were collected, pre-processed, and split into training and test sets (80:20). Random Forest and Support Vector Regression base learners were each optimised through Grid-Search cross-validation, and their tuned predictions were combined using a validation-optimised weighted hybridisation scheme. The tuned ensemble RF-SVM model achieved a Mean Squared Error (MSE) of 0.001109, a Root Mean Squared Error (RMSE) of 0.033308, and a coefficient of determination (R²) of 0.981786 on normalised test data. Predicted-versus-actual and residual plots indicate close tracking of actual demand with errors randomly distributed around zero, showing no systematic bias. Explicit hyperparameter tuning meaningfully strengthens the robustness of ensemble RF-SVM models against SME demand fluctuations, offering a practical route to more reliable, data-driven inventory and promotion planning for retail SMEs undergoing digital transformation.
Copyrights © 2026