Evaluation of Radial Basis Function Network and Supervised Machine Learning Methods on Brain Stroke Prediction Datasets

Abstract

Objective: Supervised machine learning algorithms and neural networks are widely used classification methods in data mining. In this study, RBFN, one of the widely used supervised machine learning (SML) algorithms and neural network methods, was used according to the factors affecting the diagnosis of cerebral palsy, and it was aimed to evaluate their classification performance. Material and Method: The dataset is an open source dataset, and there are a total of 4981 people with and without stroke. This dataset is modeled with RBFN from neural networks with four algorithms commonly used in supervised machine learning decision tree (DT), random forest (RF), and K-nearest neighbor (K-NN) and support machine vector (SVM). Their performance was evaluated according to performance criteria. Results: The algorithms with the highest performance according to the accuracy criteria are DT (0.954), SVM (0.954), RBFN (0.954) and RF (0.953), respectively. The K-NN algorithm was found to be higher than other methods in terms of precision (0.061) and sensitivity (0.080). Conclusion: The performances of DT, RF, SVM and RBFN methods were found to be close to each other in terms of accuracy criteria. In the deci- sion-making process, the correct classification performance of these four methods is higher than K-NN.