A Novel Method to Determine Basic Probability Assignment Based on Adaboost and Its Application in Classification.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Evidence Theory

Background:

• Determining basic probability assignment (BPA) is critical in evidence theory for accurate decision-making.
• Existing methods face challenges, particularly in scenarios with limited data.

Purpose of the Study:

• To propose a novel Adaboost-based method for calculating BPAs.
• To enhance the accuracy and effectiveness of decision-making within evidence theory frameworks.

Main Methods:

• Utilizes Adaboost to generate strong classifiers for attribute models.
• Determines singleton proposition BPAs using classification weights.
• Quantifies composite proposition BPAs via the area ratio of singleton intersections.
• Employs a recursive formula for efficient computation of area ratios.
• Combines BPAs using Dempster's rule of combination.

Main Results:

• Achieved 96.53% total recognition and 90% classification accuracy on the Iris dataset with 10% training data.
• Demonstrated effectiveness and reasonableness across various datasets.
• Showcased robust performance even with insufficient sample sizes.

Conclusions:

• The proposed Adaboost-based method provides a reliable approach for BPA determination.
• The method is effective for classification tasks, particularly under data scarcity.
• Offers a computationally efficient and accurate solution for evidence theory applications.