Application of RBF neural network in predicting thalassemia disease in Mosul city

Abstract

Thalassemia is a hereditary blood disorder that can be born to children if both parents are carriers of the gene mutation. The effect of this mutation is a faster than normal rate of destruction of red blood cells and thus iron accumulation and decreased availability of hemoglobin, The quantity, quality and shape of red blood cells are also reduced. In the current dataset, the number of severe thalassemia = 131, moderate thalassemia = 149 and 13 variables were used. A sample taken from Al-Hadbaa Specialized Hospital for Hematology and Bone Marrow Transplantation in Mosul, Iraq was used. In this study, before applying Python to this model, these variables were data cleaned to remove any gaps. The data was divided into several sections using cross-validation to test the model. Radial Basal Function (RBF) networks were used in this study to classify thalassemia patients with respect to the specified model performance measurement criteria. The experimental results revealed that RBF networks performed well with test accuracy of 96%; F1 score of 96%; high sensitivity of 97%; high specificity of 95\%; and a high positive predictive value of 95%. The resulting area under the curve was 99.5%, which is very close to the ideal for the sample. Through experiments, we found that the best setting is a learning rate of 0.1 and sixteen neurons in the hidden layer. Furthermore, a random forest model was used to identify the most significant features influencing the differentiation between types of thalassemia. The results showed that the most important features are HBA1 (adult hemoglobin) and HBF (fetal hemoglobin), which represent the main indicators for determining the type of thalassemia due to their significant impact on classification. This is followed by the HB (total hemoglobin) feature as a third important feature, and then growth delay and HBA2 with varying degrees of importance. These analyses helped identify the fundamental factors associated with the genetic and clinical differences between major thalassemia and intermediate thalassemia, contributing to enhancing the understanding of the precise classification of the disease and improving diagnostic and treatment strategies.