Prediksi Tren Penjualan Produk E-Commerce Menggunakan Teknik Analisis Big Data

Abstract

The rapid growth of e-commerce generates enormous volumes of data that can serve as a foundation for strategic decision-making through sales trend prediction. This study aims to implement Big Data analysis techniques in predicting e-commerce product sales trends on the Amazon platform and to measure the accuracy of the resulting model. The data used is a supplement product sales dataset from Kaggle, covering the period from 2020 to 2025. The method applied is the XGBoost machine learning algorithm, with preprocessing using Pandas, model evaluation using RMSE and MAPE metrics, and visualization through Streamlit. The results show that the XGBoost model successfully identified daily, weekly, and annual seasonal patterns from historical data. The system projected total sales of 502,033 units for 2025 across nine best-selling products, with Whey Protein ranked highest at 63,085 units. The MAPE value of 5.46% falls under the Highly Accurate category (< 10%), and the RMSE value was 3,673.69 units per product. A prediction accuracy rate of 98.38% further confirms the model's reliability. This study demonstrates that XGBoost-based Big Data analysis is an effective approach for predicting e-commerce sales trends and can serve as a basis for business strategy planning such as inventory management, resource allocation, and sales target setting.