Prediksi Penjualan Produk Menggunakan Algoritma Xtreme Gradient Boosting

Abstract

The rapid advancement of the digital era has intensified competition in the retail sector, including at SRC Pak Didin's store. One of the challenges faced by the store is suboptimal product management, which impacts its operational efficiency. Data Science offers solutions for enhancing business performance, such as improving operational efficiency and optimizing marketing or sales strategies. This study aims to predict product sales at SRC Pak Didin’s store using the XGBoost algorithm and to propose a sales strategy that can be applied to improve the store's operations. The dataset used in this research comprises 24 product types over the past three months, starting from September 2024. These products include instant noodles (various flavors such as fried, chicken broth, Aceh, and Soto), bottled tea, soft drinks, herbal drinks, instant coffee, snacks, biscuits, bottled water, bread, flour, sugar, seasoning mixes, ice cream, boxed milk, and other light snacks. The research employs the XGBoost algorithm to analyze sales data from the past three months and predict sales for the following month. Evaluation metrics used include Mean Squared Error (MSE) and R-squared (R²). The XGBoost model was tested in three scenarios: XGBoost Regression, XGBoost Regression Linear (single variable x), and XGBoost Regression Linear (two variables x), with the objective of identifying the best-performing model. The accuracy results show that the XGBoost Regression model achieved 96.56%, the XGBoost Regression Linear model with a single variable x achieved 99.22%, and the XGBoost Regression Linear model with two variables x achieved 99.59%. The XGBoost Regression Linear model with two variables was selected as the best model due to its highest accuracy score. This model can effectively predict product sales and provide actionable insights for developing sales strategies, benefiting SRC Pak Didin's store operations.