Classification of Diffuse Lung Diseases Using Heterogeneous Ensemble Classifiers

Abstract

Accurate classification of Diffuse Lung Diseases (DLD) plays a significant role in the identification of the lung pathology. Efficient classifiers based on various learning strategies have been proposed for multi class DLD classification. Due to imbalance in DLD class distribution the mis-classification probability of minority class is higher when compared to the majority class. To overcome the affects of imbalance in class distribution, the sampling approach is employed in the work, to balance the training set. It is observed that recognition rate of each DLD class is distinct based on the learning method adopted. Thus the complementary information offered by each classifier can be fused effectively to boost the classification performance. A heterogeneous ensemble classifier method based on weighted majority voting scheme is presented in this work to classify five DLD patterns imaged in High Resolution Computed Tomography (HRCT). The efficiency of the base and ensemble classifier is assessed based on recall, precision, F-measure and G-mean measure. By comparison it is found the results by ensemble of classifiers is superior than compared to its base classifiers.