Lung cancer stage classification on medical data with class imbalance

In this thesis, a hybrid electronic nose system comprising 8 metal oxide semiconductor sensors and 14 quartz crystal microbalance sensors was used to analyze the breath samples of individuals diagnosed with lung cancer and classify them according to disease stages. A total of 200 breath samples were analyzed, with the distribution as follows: 14 from Stage I, 12 from Stage II, 23 from Stage III, and 151 from Stage IV. Preprocessing steps such as baseline correction and conductivity transformation were applied, followed by feature extraction, Z-score normalization, and dimensionality reduction via Principal Component Analysis. Classification was performed using both the original imbalanced dataset and versions balanced with the Synthetic Minority Oversampling Technique and its variants. Logistic Regression, Support Vector Machines, k-Nearest Neighbors, and Random Forest algorithms were evaluated under different hyperparameter combinations. The performance of the models was evaluated using metrics such as accuracy, macro precision, macro recall, and macro F1-Score. The results demonstrate that both data balancing algorithms and hybrid sensor systems significantly enhance machine learning-based classification performance on imbalanced data sets.