Özet:

Blood is a vital product that thousands of people need every day due to diseases, surgeries or injuries. Therefore, it is necessary that hospitals contain sufficient amount of blood in the stock of blood banks that meet the blood needs. Less amount of blood to be obtained than necessary creates significant problems such as failure to meet the need and loss of lives, while excessive amount of blood storage leads to the damage of blood and the lack of stock of different hospitals that need blood. This study first determines the criteria affecting the demand for erythrosite suspension, which is one of the blood components; according to these criteria, it is intended to predict the demand by the network-based foolish output system (ANFIS) method adapted from machine learning algorithms. However, as there are many criteria affecting demand, the basic component analysis (PCA) method has been used to reduce by grouping and eliminate the dependencies between criteria. Furthermore, the performance of ANFIS is determined by the Taguchi experimental design method because it is associated with the correct determination of the parameters that affect the model’s structure and learning values that will provide the highest accuracy for estimation. The developed PCA-based hybrid ANFIS-Taguchi method has been applied in a region’s blood center. The correlation ratio (R) is assessed by the performance measure and the method’s predictive ability. At the end of the application, the estimated amount of demand for erythrosite suspension was found to result similar to the amount of demand that occurred at 88.1%.

Anahtar Kelimeler:

Özet:

Blood is a vital product that is needed by thousands of people every day due to diseases, surgeries or injuries. For this reason, it is necessary that the blood banks have enough blood quantity to meet the blood needs of hospitals . The provision of small amounts of blood in hospitals creates significant problems such as loss of life and can’t meet the demand. On the other hand, the stocking of large amounts of blood leads to the wastage of blood and the stockless of blood different hospitals. The aim of this study is to determine the criteria affecting blood demand and to forecast the blood demand by the machine learning algorithm Adaptive Network Based Fuzzy Inference System (ANFIS) method. However, since the number of impact criteria is high, principal component analysis (PCA) method has been used in order to decrease criteria and eliminate the dependencies between the criteria. In addition, the performance of ANFIS depend on determining ANFIS parameters that affect its structure and learning. So to provide the highest learning ANFIS parameters were determined by the Taguchi experimental design method. The developed hybrid method was applied in a regional blood center and evaluated with correlation coefficient (𝑅). At the end of the application, it is seen that the estimated red blood cells demand is similar to the demand amount realized at the rate of 88.1%.