Comparative study of artificial Neural Network and Kalman Filter models for blood demand forecasting at PMI Surabaya

Abstract

Blood plays a vital role in human health, making the need for donors and transfusions crucial. Currently, the Indonesian Red Cross (PMI) in Surabaya faces a balance issue between blood supply and demand. To address this, a blood demand forecasting model has been created at the PMI using ANN with a 4% error rate. The Kalman Filter algorithm is known to significantly reduce prediction errors from the prediction and correction process, while an ANN is considered capable of handling data complexity and nonlinearity. Therefore, this study aims to analyze the performance of the ANN and Kalman Filter models and compare the model performance results to determine the model with the best performance level. The modelling uses the CRISP-DM method, which starts from data understanding, data preparation, data modelling, model evaluation, and forecasting. The results of this study indicate that the Kalman Filter model successfully minimizes errors compared to the ANN prediction results, achieving a model accuracy level reaching 93.1%. These results demonstrate that the Kalman Filter model can significantly reduce prediction errors in the prediction and correction process, making it more optimal than the ANN model in forecasting blood demand at the PMI in Surabaya.