Improved Colony Predation Algorithm Optimized Convolutional Neural Networks for Electrocardiogram Signal...

Area of Science:

• Computational intelligence
• Swarm intelligence algorithms
• Machine learning applications

Background:

• Swarm intelligence algorithms offer flexibility for complex real-world problems.
• The Colony Predation Algorithm (CPA) is a nature-inspired algorithm with limitations in exploration and escaping local optima.
• Improving global search capability is crucial for swarm intelligence algorithms.

Purpose of the Study:

• To enhance the global search capability of the Colony Predation Algorithm (CPA).
• To propose a novel OLCPA-CNN model for parameter tuning of Convolutional Neural Networks (CNNs).
• To evaluate the performance of the improved algorithm and the proposed model on benchmark functions and real-world datasets.

Main Methods:

• An improved variant, Orthogonal Learning Colony Predation Algorithm (OLCPA), was developed by incorporating an orthogonal learning strategy.
• A novel OLCPA-CNN model was proposed, utilizing OLCPA to optimize CNN parameters.
• Comparative experiments were conducted using IEEE CEC 2017 benchmark functions and medical datasets (MIT-BIH Arrhythmia, European ST-T).

Main Results:

• The OLCPA algorithm demonstrated superior performance, ranking first compared to other traditional and advanced algorithms on benchmark functions.
• The OLCPA-CNN model achieved high classification accuracies of 97.7% and 97.8% on the MIT-BIH Arrhythmia and European ST-T datasets, respectively.
• The proposed OLCPA effectively addresses the limitations of CPA, particularly in escaping local optima.

Conclusions:

• The Orthogonal Learning Colony Predation Algorithm (OLCPA) significantly improves upon the original CPA's global search capabilities.
• The OLCPA-CNN model presents a powerful approach for accurate medical data classification.
• This research contributes to the advancement of swarm intelligence algorithms and their practical applications in machine learning.