Akademik Veri Yönetim Sistemi
Asthma is one of the most common chronic respiratory diseases worldwide, and early and accurate diagnosis is critical for effective clinical management. In this study, we evaluated the diagnostic potential of machine learning models based on voice analysis as a non-invasive approach for asthma diagnosis. Using audio samples containing seven different phonetic units, the performances of 13 different machinelearning algorithms were comprehensively analyzed. The StandardScaler and SMOTE techniques were applied in the data preprocessing stage, and a 5-fold cross-validation methodology was adopted to evaluate the models. Accuracy, F1-score, sensitivity, precision, specificity, and area under the curve (AUC) metrics were used for performance evaluation. The results demonstrate that ensemble learning approaches, particularly the stacking ensemble model, exhibit superior discriminative capacity for all phonetic units. Individual models, such as neural networks and support vector machines, also produced remarkable results, whereas simpler models were limited in terms of capturing complex patterns in audio data. This study demonstrated the promising diagnostic potential of voice analysis-based ensemble learning approaches for asthma diagnosis; however, it emphasizes the need for an optimal balance between sensitivity and specificity in clinical applications.
