Prophet–LightGBM Hybrid Model Implementation in Cafe Menu Sales Prediction

Implementasi Model Hybrid Prophet–LightGBM dalam Prediksi Penjualan Menu Kafe

Abstract

This study aims to improve the accuracy of sales forecasting for cafe menu items through the development of a hybrid model that combines the Facebook Prophet and LightGBM algorithms. This hybrid model is designed to leverage the strengths of Prophet in detecting seasonal patterns and trends, as well as the ability of LightGBM to learn from residuals that are not captured by Prophet. The dataset used is sourced from Kaggle, containing cafe menu sales data, which includes information about the menu items, the quantity sold, and the transaction dates. Model evaluation was conducted using MAE (Mean Absolute Error), MAPE (Mean Absolute Percentage Error), and RMSE (Root Mean Squared Error) metrics. According to the results, the hybrid model shows significant improvement in forecasting accuracy, with MAPE of 5.83% for one menu item (cake), MAE of 0.84, and RMSE of 0.99, indicating better accuracy compared to the single models. This study is expected to provide valuable contributions to more efficient stock management and the development of more targeted marketing strategies for the cafe industry.