An Expert System for Leukocyte Classification using Probabilistic Deep Feature Optimization via Distribution Estimation

White blood cells (WBCs) are essential for immune and inflammatory responses, and their precise classification is crucial for diagnosing and managing diseases. Although convolutional neural networks (CNNs) are effective for image classification, their high computational demands necessitate feature selection to enhance efficiency and interpretability. This study utilizes transfer learning with EfficientNet-B0 and DenseNet201 to extract features, along with a Bayesian-based feature selection method with a novel optimization mechanism to improve convergence. The reduced feature set is classified using soft voting across multiple classifiers. Tests on benchmark datasets achieved over 99% accuracy with fewer features, surpassing or matching existing methods.