A CRISP-DM Framework for Analyzing and Predicting Global E-Commerce Profitability

The rapid expansion of global e-commerce has produced massive transactional data, creating an imperative to transform raw records into strategic insight across heterogeneous regional markets. Although machine learning is widely applied to e-commerce forecasting, most studies target aggregate sales within a single market and rarely examine transaction-level profitability across regions. This study addresses that gap through a CRISP-DM-based framework integrating descriptive and predictive analytics to analyze and predict global e-commerce profitability. The objectives are to characterize multi-regional sales patterns, identify the dominant drivers of transaction profitability, and compare machine-learning models for profit prediction. Using 2,000 transactions across five regions during 2023-2025, three regression models and three classification models were developed and validated through 10-fold cross-validation. XGBoost achieved the best regression performance (R² = 0.9263; RMSE = USD 35.02), while the Decision Tree attained the highest classification accuracy (98.50%; F1 = 0.9914). Feature importance analysis showed that Shipping_Ratio, Shipping_Cost, and Discount_Percent account for 96.24% of the predictive signal, indicating that logistics economics rather than regional or segment factors dominate profitability. The study’s contribution lies in reframing regional profitability disparities as a function of shipping economics and introducing Shipping_Ratio as a transferable engineered feature for cross-border e-commerce analytics.