Blood Agglutination Curated Dataset (V1): A Quality-Controlled Binary Classification Dataset for Automated Blood Group Agglutination Analysis

Blood Agglutination Curated Dataset (V1) is a curated and quality-controlled image dataset developed for binary classification of blood agglutination reactions into two categories: • Clumped (Agglutinated)• Not Clumped (Non-agglutinated) The dataset was derived from a publicly available blood agglutination image collection and further processed through a standardized curation pipeline. The preprocessing workflow included image verification, duplicate removal, quality inspection, class organization, train/validation/test split generation, metadata creation, and dataset documentation. Dataset Composition Training Images:• Clumped: 5,790• Not Clumped: 760 Validation Images:• Clumped: 843• Not Clumped: 187 Test Images:• Clumped: 376• Not Clumped: 87 Total Images:8,043 Included Files • Train / Validation / Test folders• Dataset Card• README• Dataset statistics• Distribution charts• Sample image grid• Preprocessing notebook• Citation file This dataset is intended for research in computer vision, machine learning, trustworthy AI, uncertainty estimation, calibration, and automated blood typing systems. Important Note The original medical images originate from a publicly available Roboflow Universe dataset. This release represents a curated derivative dataset produced through preprocessing, quality control, dataset restructuring, documentation, and reproducibility improvements. Users should acknowledge both the original dataset creators and this curated release where appropriate.