<![CDATA[Plaza Banten, an MSME marketplace in Banten Province, generates ordering and sales transaction data that can be leveraged to support operational decisions, particularly inventory planning and promotional timing. However, decision-making is often reactive because demand forecasting has not been systematically developed from historical transactions. This study proposes an end-to-end pipeline that transforms Plaza Banten transaction records into daily demand time-series data at the product-category (Group) level, following data preparation and modeling stages in a data mining framework. The study uses transaction data from January to December 2024 and is positioned as a continuation of a previous Market Basket Analysis (MBA) study, which indicated that high transaction volumes were dominated by packaged rice products (e.g., rice boxes and chicken rice packages), motivating a forecasting follow-up for high-demand categories with recurring purchase patterns. The preprocessing stage includes data cleaning, validation of quantity and unit price, feature construction (quantity and revenue), daily demand aggregation by category, and completion of missing calendar dates to form continuous time series. For modeling, this study compares baseline forecasting methods (Naïve and 7-day Moving Average) against an Exponential Smoothing (Holt–Winters/ETS) model that accounts for trend and weekly seasonality. Model performance is evaluated using MAE, RMSE, and MAPE to ensure measurable selection of the best approach. The forecasting results are then interpreted as operational insights to estimate demand levels per category and support inventory planning and promotional prioritization based on predicted demand trends.]]>
Copyrights © 2026
