Grammar-based automatic programming for medical data classification: an experimental study

• Published in: The Artificial intelligence review Vol. 54; no. 6; pp. 4097 - 4135
• Main Authors: Si, Tapas, Miranda, Péricles, Galdino, João Victor, Nascimento, André
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 01.08.2021; Springer; Springer Nature B.V
• Subjects: Algorithms; Artificial Intelligence; Artificial neural networks; Classification; Classifiers; Computer Science; Decision analysis; Decision-making; Diagnosis; Discriminant analysis; Feature extraction; Geometric accuracy; Grammar, Comparative and general; Machine learning; Medical advice systems; Multiple criterion; Performance evaluation; Grammatical swarm; Automatic programming; Medical data classification; Swarm programming
• ISSN: 0269-2821; 1573-7462
• DOI: 10.1007/s10462-020-09949-9

In a computational medical model, diagnosis is the classification of disease status in the terms of abnormal or positive , normal or negative or intermediate stages . Different Machine learning techniques such as artificial neural networks (ANNs) are extensively and successfully used in disease diagnosis. However, there is no single classifier that can solve all classification problems. Selecting an optimal classifier for a problem is difficult, and it has become a relevant subject in the area. This paper focuses on grammar-based automatic programming (GAP) to build optimized discriminant functions for medical data classification in any arbitrary language. These techniques have an implicit power of automatic feature selection and feature extraction. This work carries out an in-depth investigation of the use of different GAP algorithms in the medical data classification problem. The objective is to identify the benefits and limitations of algorithms of this nature in the current problem. Classical classifiers were also considered for comparison purposes. Fourteen medical data sets were used in the experiments, and seven performance measures such as accuracy, sensitivity, specificity, precision, geometric-mean, F-measure, and false-positive rate are used to evaluate the performance of the produced classifier. The multiple criteria decision analysis (MCDA) demonstrates that GAP approaches are able to produce suitable classifiers for a given problem, and the GS performs better than other classical classifiers in medical data classification.