Feature selection algorithms to improve thyroid disease diagnosis

Correct and early diagnosis of diseases is important and mandatory in healthcare industry for correct and timely treatment. This fact is more important in diseases like thyroid, which is very difficult to detect as its symptoms coincide with several diseases. Usage of machine learning algorithms for thyroid disease diagnosis is prominent. A typical thyroid disease diagnosis system uses three main steps, namely, feature extraction, feature selection and classification. The main goal of this paper is to analyze the use of filter-based (F-Score) and wrapper-based (Recursive Feature Elimination) feature selection algorithms on its effect on disease identification and classification. The analysis is also performed with Principle Component Analysis dimensionality reduction algorithms. Performance evaluation was performed with three metrics, namely, accuracy, sensitivity and specificity. Four classifiers, namely, MultiLayer Perceptron, Back Propagation Neural Network, Support Vector Machine and Extreme Learning Machine were used to analyze the selected algorithms. Experimental results showed that while both F-Score and Recursive Feature Elimination improved the performance of thyroid disease diagnosis, the wrapper-based algorithm produced maximum efficiency and produced a maximum accuracy of 98.14% with ELM classifier.