Classification of Lung Diseases Through Artificial Intelligence Models: A Multi-Dataset Evaluation

Respiratory ailments like chronic obstructive pulmonary disease(COPD), asthma, and pulmonary fibrosis pose substantial diagnostic challenges that impact global health. This study combines artificial intelligence with non-invasive diagnostics, employing models from machine learning and deep learning model to evaluate on three open source lung sounds datasets. Mel-frequency cepstral coefficients (MFCCs) was employed for feature extraction and a range of models were evaluated including Support Vector Machines (SVMs), Random Forests (RFs), Gradient Boosting, Naive Bayes (NB), Artificial Neural Network (ANN), Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks and a transformer model for classifying respiratory sounds into nuanced categories, with defined performance metrics. Results illustrate the criticality of selecting models tailored to dataset characteristics; where prominent models include SVM, LSTM, and CNNs, to detect disease-specific acoustic patterns. This offers a path to understand toward accessible and efficient diagnostic methods in relation to open source datasets.