Optimization-Based Hyperparameter Selection in Deep Learning Methods for Detection of Lung Diseases

Lung diseases are a major health problem today. Lung diseases caused by various factors, especially air pollution, smoking habits, respiratory tract infections, and genetic factors, negatively affect the quality of life of individuals and can cause serious health problems. This study addresses the analysis of audio files for the detection of lung diseases. The data obtained from a total of 236 audio files were first converted into 4096 features using the VGG16 pre-trained transfer learning model. Then, these features were classified with Support Vector Machines (SVM) and an accuracy of $\mathbf{9 2 . 4 0 \%}$ was achieved. In order to increase the performance of the SVM model, hyperparameter optimization was performed with the Salp Swarm Algorithm (SSA) and the accuracy rate was increased to $\mathbf{9 6 . 6 1 \%}$. The results of the study show that feature extraction with VGG16 and optimization with SSA is an effective method in detecting lung diseases. This research emphasizes that combining transfer learning and metaheuristic optimization techniques can make a significant contribution to achieving higher accuracy rates in medical diagnostic processes. This study demonstrates the successful use of artificial intelligence techniques in important medical applications such as effective analysis of voice data and detection of lung diseases.