<![CDATA[Machine learning models in the healthcare domain are capable of achieving high accuracy; however, they often function as black-box systems, making them less transparent and potentially introducing bias toward sensitive groups. This study aims to analyze the performance, interpretability, and fairness of disease classification models using five tabular medical datasets, namely Alzheimer, Obesity, Hypertension, Stroke, and Asthma datasets obtained from Kaggle. The research stages include data cleaning, feature transformation, normalization, and handling class imbalance using SMOTE. The models were developed using Support Vector Machine (SVM) and Random Forest algorithms with hyperparameter optimization through GridSearchCV and validation using 5-fold cross-validation. The results indicate that Random Forest provided the most consistent performance, achieving the highest accuracy of 96.92% on the Obesity dataset. In imbalanced datasets such as Stroke and Asthma, model performance declined, particularly in terms of precision and F1-score, due to uneven class distribution and data complexity. Interpretability analysis using SHAP and LIME demonstrated that the models utilized clinically relevant features, such as age, blood pressure, body mass index, and cognitive function indicators. Fairness evaluation using Demographic Parity Difference (DPD) and Equal Opportunity Difference (EOD) produced relatively small values, indicating that the distribution of predictions across sensitive groups, particularly gender, was fairly balanced, although still influenced by data characteristics. This study confirms that integrating performance, interpretability, and fairness in multi-dataset evaluation provides a more comprehensive approach compared to conventional evaluations that focus solely on accuracy.]]>
Copyrights © 2026
