Multiclass Alpha Integration of Scores from Multiple Classifiers.

Area of Science:

• Machine Learning
• Signal Processing
• Biomedical Engineering

Background:

• Alpha integration methods are established for binary classification tasks.
• Multiclass classification is prevalent in automatic classification problems.
• Score fusion techniques are crucial for improving classification performance.

Purpose of the Study:

• Introduce Separated Score Integration (SSI), a novel alpha integration-based method for soft score fusion in multiclass classification.
• Theoretically derive parameter optimization for SSI to minimize mean squared error (LMSE) or minimum probability of error (MPE).
• Evaluate SSI's performance against single classifiers and classical fusion techniques.

Main Methods:

• Developed Separated Score Integration (SSI) based on alpha integration for soft fusion in multiclass problems.
• Optimized SSI parameters using theoretical derivation for LMSE and MPE.
• Validated SSI on simulated ultrasonic pulse classification (4-class), polysomnographic sleep staging (3-class), and electroencephalographic seizure detection (5-class).

Main Results:

• SSI demonstrated superior performance compared to individual classifiers across all tested datasets.
• The proposed method consistently outperformed classical fusion techniques.
• Experiments included simulated ultrasonic pulses, real polysomnographic sleep data, and real electroencephalographic data.

Conclusions:

• Separated Score Integration (SSI) offers an effective approach for soft score fusion in multiclass classification.
• SSI provides a significant performance improvement over existing methods, including single classifiers and classical fusion techniques.
• The method's efficacy is validated across diverse real-world and simulated classification challenges.