Predicting the Behavior of Blood Donors in Tehran Blood Transfusion Center Using Data Mining Techniques

Blood donation is a vital public health activity and accurately identifying eligible donors is crucial for ensuring a safe and sufficient blood supply chain. Understanding donor behavior and characteristics helps optimize recruitment strategies and enhance donor management. This study analyzed ۵,۰۰۰ anonymized records from the Tehran Blood Transfusion Center, including ۲۲ demographic, behavioral and medical features. Descriptive analyses revealed near-normal distributions for physiological metrics such as weight, blood pressure, hemoglobin, hematocrit, pulse, temperature and cholesterol, while age, lifetime donations and days since last donation showed skewed or multimodal patterns. Hematological factors, particularly hemoglobin (۰.۴۸۶) and hematocrit (۰.۴۱۹), were strongly associated with donor eligibility, whereas smoking had a minor negative effect. Several classification models were evaluated using accuracy, precision, recall, specificity and F۱-score. Ensemble methods, notably Gradient Boosting and XGBoost, outperformed traditional models. Gradient Boosting achieved the highest accuracy (۰.۹۷۴) and F۱-score (۰.۹۸۳), while ROC curve analyses and AUC values further confirmed the superior discriminative ability of these models in distinguishing eligible from ineligible donors.

کلیدواژه ها:

Blood donation ، Donor eligibility ، Classification model ، ROC curve ، AUC