Comparative Analysis of Deep Learning LSTM and Prophet Models in Predicting Victori Self Service Sales Trends

Abstract

Sales forecasting is a crucial aspect of operational planning in the retail business, as it helps companies manage inventory, design promotional strategies, and optimize supply chains. This study compares the performance of two time series forecasting methods, namely Long Short-Term Memory (LSTM) and Prophet, in predicting daily sales trends at Victory Swalayan during the period from March 1 to May 30, 2025. The dataset consists of daily transaction records that were aggregated into a daily sales time series. The LSTM model was trained using a multi-step iterative approach with a seven-day input window, while the Prophet model was built using default settings with weekly seasonality and trend components. Model performance was evaluated using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) metrics. The results show that the Prophet model achieved higher accuracy than LSTM, with an MAE of 1,342,401 and an RMSE of 1,615,108, while LSTM recorded an MAE of 1,986,710 and an RMSE of 2,502,594. Therefore, Prophet is more effective in modeling seasonal patterns and daily sales trends in retail business data.