Stacked ensemble model based pulmonary abnormality detection: Improved chi-square feature selection including Wavelet...

Pulmonary abnormality detection via analysis on respiratory sound is a novel method that shows promising results in early identification and detection of respiratory disorders. Researchers and physicians can benefit greatly from the Respiratory Sound Database that includes recordings of different respiratory sounds belonging to both healthy and unhealthy lung activities. By utilizing the latest developments in signal processing and ML methodologies, this work aims to create a new method termed as stacked ensemble model based Pulmonary abnormality detection (SEM based PAD) that can automatically identify and categorize abnormalities from the recordings of sounds. This article deploys Improved Wiener filtering (IWF) for preprocessing the sound signal. Subsequently, extraction of varied features takes place with improvement in Statistical parameter determination (ISPD). Then, Improved chi-square is used to choose the features, thereby, lessening the length of features. Finally, stacked ensemble model (SEM) that combines Squeeze Net, SVM and CNN is deployed for detecting the pulmonary abnormalities. The final outcomes are determined using Improved Score Level Fusion (ISLF).