<![CDATA[Accurate demand forecasting is vital in supply chain management, particularly in the fast-moving consumer goods (FMCG) industry that experiences rapid stock turnover and fluctuating demand. The Auto-Regressive Moving Average (ARMA) has been the standard approach for time series forecasting, however it often underperforms under sparse and fluctuating data. This study contributes to the literature by applying Demand Response-ARMA (DR-ARMA) that was initially developed to address data sparsity and fluctuations under more complete and lower-sparsity data conditions. Using three primary datasets with varying sparsity levels from an FMCG distributor of bottled water products in West Borneo, DR-ARMA was benchmarked against classical ARMA. The results show that DR-ARMA consistently outperforms the classical ARMA model even under more complete, lower sparse data conditions. In lower sparsity datasets, DR-ARMA achieved average Mean of Percentage Error (MAPE) values of 22.64% and 6.41% respectively compared to the baseline ARMA model (235.60% and 180.86%). However, its best performance was observed in higher sparse condition (70.45%), achieving an average MAPE value of 1.79% across all datasets, suggesting the model remains most effective when applied to sparse data as originally intended. These improvement enables more precise safety stock planning, lower holding costs, and position DR-ARMA as a practical forecasting tool that connects analytical performance with real operational impact.]]>
