Raw Material Stock Prediction Using the Long Short-Term Memory Algorithm

Sari

Inaccurate management of raw material inventory leads to operational inefficiency and cost overruns in micro, small, and medium enterprises (MSMEs), particularly in the culinary industry where demand is highly fluctuating and difficult to predict. This study develops a raw material stock prediction system using the Long Short-Term Memory (LSTM) algorithm with a Waterfall system development approach, applied to the case of "Mizan and Sunan" grilled bread producers operating across seven branches. The dataset consists of nine months of historical demand data, comprising 5,142 entries with eight main attributes. Data preprocessing includes Min-Max Scaling normalization, sequential data formation using a three-day sliding window, and chronological splitting of training and testing datasets. The LSTM model is trained to predict daily stock requirements, with evaluation conducted using Mean Squared Error (MSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). The results show an MSE of 403.28, MAE of 10.38, and MAPE of 10.79%, indicating good predictive accuracy. The novelty of this research lies in the application of an LSTM model based on multi-branch MSME culinary historical data characterized by fluctuating demand, along with the development of an adaptive prediction system to support precise procurement decision-making. These findings demonstrate the effectiveness of LSTM as a practical data-driven solution for inventory management in multi-branch MSME operations.