Anemia is the most common blood disorder worldwide. It requires many blood tests, radiological images, and various diagnostic tests for accurate diagnosis. It is also important to ensure that patients receive the most appropriate and effective treatment. Given the rising number of patients and the challenges in accessing medical specialists, this study proposes a robust system for the classification of anemia types using complete blood count (CBC) data, utilizing four Artificial Neural Network Models: Residual Network (ResNet)-Based Model, Recurrent Neural Networks (RNN), Multilayer Perceptron (MLP), and Convolutional Neural Networks (CNN). The study evaluates a dataset comprising 15,300 samples across five distinct anemia classes: Non anemia records, Hgb-anemia, folate deficiency anemia (FDA), iron deficiency anemia (IDA), and B12 deficiency anemia. Data normalization was performed to address the substantial numerical differences in parameter values. Additionally, the Synthetic Minority Over-sampling Technique (SMOTE) was employed to rectify class imbalances within the dataset. Each model is validated using Precision, Recall, F-score, and Accuracy values. The results demonstrated that ResNet achieved the highest accuracy of 97.34%, closely followed by RNN at 97.31%, MLP at 96.45%, and CNN at 93.82%. The ResNet and RNN models exhibited exceptional performance across both majority and minority classes, indicating their effectiveness in generalizing under imbalanced conditions. This study emphasizes the potential of artificial intelligence in clinical decision-making, providing significant insights for healthcare professionals in diagnosing various anemia types efficiently.