Fashion Product Sales Prediction Using Random Forest Regression Algorithm Based on Historical Data

Abstract

Tulip Store at Metropolitan Mall Bekasi, managed by PT Star Asia Brothers,relies on historical sales data for evaluation and target planning. However,sales volume fluctuations, ranging from 385 units in October 2025 to1,122 units in March 2026, make heuristic planning unreliable. This studybuilds a daily sales volume prediction model using Random Forest Regressionbased on daily recapitulation data from October 2025 to March 2026. Followingthe CRISP-DM framework, data preparation involved cleaning, dailyaggregation, and feature engineering to construct 11 input features categorizedinto temporal indicators, historical lag variables, and rolling statistics.The model was trained using an 80:20 train-test split with hyperparametersn_estimators = 200, max_depth = None, and random_state = 42. Evaluationon 36 test instances yielded a Mean Absolute Error (MAE) of 1.7649and a Root Mean Square Error (RMSE) of 2.9740, with 91.67% of predictionsexhibiting errors within ±5 units. Feature importance analysis revealed thatsequential month (Bulan_ke, 0.385) and previous-day sales (Lag_1, 0.294)were the most influential variables, indicating that monthly seasonal trendsand immediate sales momentum drive daily demand. The proposed modelprovides robust operational decision support for daily sales evaluation, targetsetting, and stock allocation.