Model Selection for Body Temperature Signal Classification Using Both Amplitude and Ordinality-Based Entropy Measures

Area of Science:

• Signal processing
• Biomedical engineering
• Statistical modeling

Background:

• Entropy-related methods are widely used for signal classification.
• Optimal selection of entropy measures and parameters can be challenging, leading to suboptimal performance.
• Existing methods struggle to achieve high classification accuracy for body temperature records.

Purpose of the Study:

• To enhance signal classification accuracy in suboptimal situations.
• To investigate the synergistic effects of combining different entropy measures.
• To develop improved statistical models for signal classification.

Main Methods:

• Utilized permutation entropy (PE), approximate entropy (ApEn), and sample entropy (SampEn).
• Developed statistical models using uncorrelated entropy measures to exploit potential synergies.
• Applied a logistic model to combine PE and ApEn for classification.

Main Results:

• Individual entropy measures yielded classification accuracy below 80% for body temperature records.
• Combining PE and ApEn with a logistic model increased classification accuracy to 90%.
• The proposed combined approach demonstrated superior performance compared to individual measures.

Conclusions:

• Combining different types of entropy measures (ordinal vs. amplitude-based) can enhance classification accuracy.
• Statistical models leveraging synergistic effects of uncorrelated measures offer a promising approach for signal classification.
• The developed method significantly improves the classification of body temperature records.