Enhanced multiclass cardiovascular disease detection via Kernel-PSO feature selection: a comparative benchmark of gradient-driven and bio-inspired machine learning algorithms
Swarm and Evolutionary Computation, cilt.105, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 105
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.swevo.2026.102405
- Dergi Adı: Swarm and Evolutionary Computation
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, INSPEC
- Anahtar Kelimeler: Batch least squares, Feature selection, Gradient-driven learning, Kernel-particle swarm optimization, Multiclass cardiovascular detection
- Sağlık Bilimleri Üniversitesi Adresli: Hayır
Özet
The global burden of the cardiovascular disease is compounded by a critical diagnostic gap, where the traditional electrocardiogram (ECG) interpretation often fails to resolve overlapping multiclass feature spaces. This paper addresses these inconsistencies by proposing feature selection-based robust machine learning algorithms for the automated detection of the Arrhythmia (ARR), Congestive Heart Failure (CHF), Atrial Fibrillation (AF), and Normal Sinus Rhythm (NSR). By introducing a performance-driven Kernel-Particle Swarm Optimization (Kernel-PSO) approach with a polynomial basis function, the research replaces the traditional accuracy-centric metrics with an objective function that eliminates the empirical trial-and-error process and provides a precise mathematical basis for the optimal features to detect multiclass diseases. Experimental results demonstrate that the objective function decreases monotonically as the feature count increases, consistently identifying the most informative features across various subsets. A rigorous benchmarking analysis reveals that the proposed feature selection-based Batch Least Squares (BLS) and Non-negative Least Squares (NNLS) significantly outperform the bio-inspired Firefly machine learning algorithm in terms of convergence stability and testing robustness. Specifically, the BLS provides quantified proof of superior performance when evaluated with the unseen data, achieving a F1-score of 0.98, a significantly low RMSE of 0.04, and a superior Cohen’s Kappa value of 0.96, thereby ensuring high diagnostic reliability and generalization capability across the multiclass dataset.