A Multi-Objective Genetic Algorithm-Optimized LightGBM Framework for Customer Segmentation and Strategy Optimization in Cross-Border E-Commerce

Abstract

Cross-border e-commerce has grown rapidly in the context of changing global trade, necessitating the urgent need for more clever and flexible consumer segmentation techniques to improve strategic operations and precision marketing. When it comes to managing high-dimensional, noisy, and behaviourally varied client data, traditional segmentation strategies often fail. In order to bridge this gap, this work suggests a unique hybrid technique that optimizes client segmentation and personalised strategy suggestion by combining the Light Gradient Boosting Machine (LightGBM) with a Selective Multi-Objective Genetic Algorithm (MOGA). To address concerns of incompleteness, duplication, and inconsistency, the technique entails gathering customer contact, order, and product data from a cross-border cosmetics e-commerce platform. Based on the data of 7000 customer purchases on a cosmetics online site. They are frequency, monetary value, product types and time-purchase trends. The data were divided into 70 percent training data, 15 percent validation and 15 percent testing data after pre processing. The proposed MOGA-LightGBM model aims at optimizing the accuracy, F1-score, and ROI by simultaneously tuning the hyper parameters and feature selection. This is followed by a thorough preparation of the data. The suggested MOGA-LightGBM model optimizes a number of factors, including accuracy, recall, and campaign efficiency, in order to maximize classification performance and marketing ROI at the same time. According to experimental data, the model performs better than benchmark methods (RFM + K-Means, XGBoost) with an accuracy of 93%, an F1-score of 85%, and a strategy ROI improvement of 17.4%. This study offers a scalable, data-driven framework for precise operations in cross-border e-commerce and highlights the possibilities of evolutionary optimization in consumer analytics.