Feature Interaction-Based Blood Cell Classification: A Comparative Study of Machine Learning and Ensemble Methods
2026 34th Signal Processing and Communications Applications Conference (SIU), İstanbul, Türkiye, 7 - 10 Temmuz 2026, ss.1-4, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/siu71813.2026.11637060
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.1-4
- Ondokuz Mayıs Üniversitesi Adresli: Evet
Özet
Accurate and efficient classification of blood cells is crucial for the early diagnosis of hematological diseases. In this study, a feature engineering-based approach is proposed for multi-class classification of blood cells by leveraging morphological and color-based feature interactions. To enhance the representational power of the tabular dataset, meaningful interactions are constructed among features such as cell area, perimeter, color channels, and staining intensity. Various machine learning algorithms are evaluated on the proposed feature set, with a particular focus on comparing gradient boosting-based methods with conventional classifiers. In addition, ensemble learning approaches, including weighted voting and stacking, are investigated to assess their potential performance improvements. Experimental results have shown that the CatBoost model outperformed all other methods with 98.55% accuracy and a macro F1 score of 97.92%. In contrast, it has been observed that ensemble methods have not been able to significantly surpass this performance level.