RedAid: A Web-Based Blood Bank Management System with Machine-Learning Support for Request Prioritization and Donor Matching

Many blood banks in Nepal still keep track of donors, requests and stock with phone calls, registers and spreadsheets. As a result, an emergency request can sit in the same queue as a routine one, and a suitable donor is often found too late. Commercial systems such as eBloodBank and BloodTrack have digitized inventory and scheduling, but neither reads the reason written on a request nor tries to spot donors who are drifting away. This paper describes RedAid, a web application built with React, Laravel and MySQL, backed by a small Python FastAPI service for its learning components. RedAid labels each request as High, Medium or Low priority using multinomial Naive Bayes together with an emergency-keyword rule, ranks nearby donors by Haversine distance, estimates donor churn with logistic regression, forecasts daily demand with Prophet and groups past campaign sites with K-Means. All 26 functional test cases passed. When we evaluated the priority classifier with leave-one-out cross-validation on its 18 training sentences, however, it reached only 38.9% accuracy (macro F1 of 0.33), and the keyword rule marked negated phrases such as "Not an emergency" as High. We present the design and implementation, report these results as measured, and identify a larger labelled dataset and real hospital records as the most important next steps.