Imbalanced Multiclass Medical Data Classification based on Learning Automata and Neural Network

Data classification in the real world is often faced with the challenge of data imbalance, where there is a significant difference in the number of instances among different classes. Dealing with imbalanced data is recognized as a challenging problem in data mining, as it involves identifying minority-class data with a high number of errors. Therefore, the selection of unique and appropriate features for classifying data with smaller classes poses a fundamental challenge in this research. Nowadays, due to the widespread presence of imbalanced medical data in many real-world problems, the processing of such data has gained attention from researchers. The objective of this research is to propose a method for classifying imbalanced medical data. In this paper, the hypothyroidism dataset from the UCI repository is used. In the feature selection stage, a support vector machine algorithm is used as a cost function, and the wrapper algorithm is employed as a search strategy to achieve an optimal subset of features. The proposed method achieves high accuracy, reaching 99.6% accuracy for data classification through the optimization of a neural network using learning automata.