Enhanced K-means algorithm for user segmentation and precision marketing strategy in E-commerce platforms

User profiles in multi-touchpoint e-commerce scenarios are high-dimensional, sparse, heavy-tailed, and time-varying. Traditional clustering methods are sensitive to initialization and noise, making them difficult to maintain stability and interpretability across time periods. To address this, this paper constructs a unified framework: PCA dimensionality reduction, adaptive K selection driven by combined cluster validity (CVI), and density-constrained K-Means. Evaluated on 6-month data of 1 million e-commerce users, this framework maintains near-linear complexity and enhances robustness to non-uniform distributions, our approach outperforms various baselines (Silhouette = 0.463, Davies–Bouldin = 0.51, Bootstrap-ARI = 0.83) and achieves stable business gains (CVR +6.8%, AOV +4.1%, RPU +6.4%, and ROI +9.7%) over a 28-day online comparison. Results demonstrate that this framework achieves a quantifiable balance between quality, stability, and efficiency.