Enhancing blockchain transaction classification with ensemble learning approaches.

Area of Science:

• Computer Science
• Data Science
• Cybersecurity

Background:

• Blockchain technology offers secure information sharing across finance, SCM, and IoT.
• Growing user numbers necessitate robust methods for identifying malicious blockchain transactions.
• Maintaining trust in blockchain ecosystems requires accurate transaction risk assessment.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) model for classifying blockchain transactions as risky or non-risky.
• To explore the efficacy of ensemble feature selection and classification methods for this task.

Main Methods:

• Employed four feature selection techniques: CFS, RFE, RF, and IG.
• Utilized ensemble feature selection (rank averaging and rank aggregation) to combine feature subsets.
• Applied ensemble classification (hard voting, soft voting, weighted averaging) to base learner predictions.
• Evaluated the model on three distinct blockchain transactional datasets.

Main Results:

• The Rank Averaging ensemble feature selection achieved a maximum accuracy of 99.24%.
• The Rank Aggregation ensemble feature selection reached a maximum accuracy of 98.73%.
• The proposed ensemble-based model demonstrated high performance in classifying transaction risk.

Conclusions:

• Ensemble methods, particularly Rank Averaging, are highly effective for feature selection in blockchain transaction risk classification.
• The developed ML model significantly enhances the ability to identify risky transactions, bolstering trust in blockchain networks.
• This research provides a scalable and accurate solution for securing blockchain ecosystems against malicious activities.