Feature selection in pulmonary function test data with machine learning methods

Pulmonary function test has vital importance in diagnosis and treatment of lung diseases. With this test, several parameters are measured such as forced vital capacity (FVC) and forced expiratory volume in the first second (FEV₁) of patients. These parameters indicate different types of lung disorders. Main constraint in diagnosis is to selection of important parameters among test results. In this study, five results of pulmonary function test (PFT) are evaluated with machine learning methods and feature selections with test results are achieved. Feature selections are performed with using Naive bayes, support vector machine (SVM), linear discriminant analysis (LDA) and k-nearest neighbor classifier (k-NN) methods. The test results of 436 patients are obtained from Atatürk Chest Diseases and Thoracic Surgery Training and Research Hospital in Ankara/Turkey. SVM method has a highest performance values with 89,6% accuracy, 87,4 % specificity, 71,6% sensitivity respectively. Thus, it is found with feature selection that importance order of test results are FVC, FEV₁, FEV₁/FVC, PEF ve FEF_25/75 respectively. In this study, obtained performance values are higher than most of studies in the literature.