Ensemble Decision for Spam Detection Using Term Space Partition Approach.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Spam detection is crucial for email security.
• Existing methods often struggle with effectively utilizing both global and local email features.
• The Term Space Partition (TSP) method offers a novel way to manage feature spaces.

Purpose of the Study:

• To propose an ensemble decision approach for effective spam detection.
• To combine global and local email features using an extended TSP method.
• To enhance the performance, efficiency, and robustness of spam detection systems.

Main Methods:

• Utilized Term Space Partition (TSP) to divide term space into subspaces with distinct feature construction strategies.
• Extended TSP with a sliding window technique to extract local email features.
• Constructed global and local classifiers, combining their decisions via a voting technique for ensemble decision-making.

Main Results:

• The proposed ensemble method demonstrated significant performance improvements over the base TSP approach.
• Experimental results on five benchmark corpora confirmed enhanced robustness.
• The method outperformed prevalent and state-of-the-art spam detection approaches.

Conclusions:

• The ensemble spam detection approach effectively integrates global and local features.
• The method offers superior performance, efficiency, and robustness compared to existing techniques.
• This approach exhibits flexible capability and adaptivity for real-world spam filtering applications.