Weighted Random Forests to Improve Arrhythmia Classification.

Area of Science:

• Machine Learning
• Cardiology
• Signal Processing

Background:

• Ensemble models combine diverse predictive learners, raising questions about optimal weighting and parameter tuning.
• Weighted ensembles often outperform simple averaging, necessitating advanced weighting strategies.
• Accurate detection of cardiac arrhythmias is crucial for intensive care patient management.

Purpose of the Study:

• To propose a new weighting algorithm for Random Forest models.
• To comprehensively examine optimal parameter tuning for the proposed weighting scheme.
• To evaluate the algorithm's performance on real-world intensive care patient data.

Main Methods:

• Development of a novel tree-weighting algorithm for Random Forest models.
• Application and evaluation of the algorithm on the Physionet/Computing in Cardiology Challenge 2015 dataset.
• Classification of cardiac arrhythmias based on electrocardiogram and pulsatory waveform signals.

Main Results:

• The proposed weighting approach demonstrated improved classification accuracy for three challenging cardiac arrhythmias.
• The algorithm showed flexibility, good performance, stability, and resistance to overfitting.
• Performance was evaluated against the standard Random Forest model on critical care data.

Conclusions:

• The novel weighting algorithm offers a significant improvement in Random Forest-based cardiac arrhythmia detection.
• The approach provides a robust and effective method for analyzing intensive care patient signals.
• This work contributes to advancing automated diagnostic tools in critical care cardiology.