COMPARING OF ARTIFICIAL NEURAL NETWORK AND MULTIPLICATIVE HOLT WINTERS EXPONENTIAL SMOOTHING METHODS IN FORECASTING DEMAND

General Background: Fluctuations in raw material orders pose significant challenges for business operators, especially during peak seasons like holidays and the new year, often resulting in shortages or excess inventory. Specific Background: This study focuses on forecasting demand for wallet products from UMKM Pengerajin Dompet Khas Tanggulangin (PDKT) by comparing two forecasting methods: Artificial Neural Networks (ANN) and the Multiplicative Holt-Winters method, which is tailored for seasonal data. Knowledge Gap: While existing literature recognizes the effectiveness of various forecasting techniques, there is limited comparative analysis of ANN and Holt-Winters specifically in the context of UMKM wallet production, highlighting the need for empirical validation. Aims: This research aims to identify the most accurate forecasting method to optimize raw material usage and production planning. Results: The findings indicate that the ANN method yields a superior Root Mean Square Error (RMSE) of 14.249, compared to 93.436 for the Holt-Winters method, establishing its higher predictive accuracy. Novelty: The study contributes to the field by providing a comparative analysis of forecasting methods tailored to the specific context of UMKM, demonstrating the efficacy of ANN over traditional methods. Implications: These results suggest that adopting ANN for demand forecasting can significantly enhance inventory management and production efficiency for PDKT MSMEs, ultimately leading to better resource allocation and reduced operational costs.