SE-stacking: Improving user purchase behavior prediction by information fusion and ensemble learning.

Area of Science:

• Computer Science
• Machine Learning
• E-commerce Analytics

Background:

• Online shopping behavior presents challenges due to data sparsity and high dimensionality.
• Previous user behavior prediction models often overlooked crucial aspects like feature selection and ensemble design.

Purpose of the Study:

• To propose an advanced SE-stacking model for accurate user purchase behavior prediction in e-commerce.
• To improve machine learning algorithm performance by focusing on feature selection and ensemble strategies.

Main Methods:

• Implemented an ensemble feature selection method to identify key purchase-related factors.
• Utilized a stacking algorithm with ten diverse base learners, optimized for prediction accuracy.
• Applied information fusion and ensemble learning principles within the SE-stacking framework.

Main Results:

• The SE-stacking model achieved a high F1 score of 98.40% on a public dataset.
• Demonstrated a performance improvement of approximately 0.09% over the best-performing base models.
• Validated the model's effectiveness in predicting user purchase behavior.

Conclusions:

• The proposed SE-stacking model offers significant practical value for e-commerce applications.
• This approach holds academic importance for advancing user behavior prediction and machine learning in e-commerce.
• The model's robust performance highlights the benefits of integrating ensemble learning and feature selection.