Prediction of Acute Myeloid Leukemia Subtypes Based on Artificial Neural Network and Adaptive Neuro-Fuzzy Inference System Approaches

The proposed technique involves designing and implementing an acute myeloid leukemia sub-type prediction system based on artificial neural network and adaptive neuro-fuzzy inference system approaches. The dataset of 600 possible cases (patients) of acute myeloid leukemia is used. After training the system with 540 input-output dataset of patients having AML-M0, AML-M1, AML-M2, AML-M3, and AML-M4 types of leukemia, it is tasted with 60 data for validation. The method is implemented to predict these five types of acute myeloid leukemia based on the characteristics of four complete blood count (CBC) parameters, namely leukocytes, hemoglobin, platelets, and blasts of the patients. The neural network performed well than the adaptive neuro-fuzzy inference system when test data was considered, where the average mean squared error (MSE) for each system was 0.0433 and 0.2089, respectively. The adaptive neuro-fuzzy inference system showed better performance than artificial neural network when training data was considered, where the mean squared error (MSE) for each system was 0.0017 and 0.0044, respectively.