Research Information System
Journal of Artificial Intelligence with Applications, vol.7, no.1, pp.1-6, 2026 (Peer-Reviewed Journal)
Malignant pleural mesothelioma (MPM) is a highly aggressive cancer with delayed diagnosis due to nonspecific symptoms and the need for invasive procedures. Therefore, noninvasive diagnostic strategies using readily available blood and clinical parameters are urgently needed to early and effective diagnosis of MPM. To overcome the limitations of manual interpretation, this study proposes a machine learning-based framework for automated MPM classification using demographic data and laboratory values. A publicly available Mesothelioma dataset from UCI Machine Learning Repository was used, including normal and diseased demographic data and laboratory values. Firstly, the Mesothelioma dataset was preprocessed, and two variables were removed during exploratory data analysis due to data leakage arising from their near-perfect correlation with the outcome, which could bias model learning. Then, in the automated MDM classification stage, several machine learning models were tested and compared, including Logistic Regression (standard and L1-regularized), Ridge Classifier, Support Vector Machines (with linear and RBF kernels), Random Forest, and XGBoost. Among the evaluated models, XGBoost achieved the highest accuracy of 73.00%, while two machine learning algorithms-namely L1-regularized Logistic Regression and Linear Support Vector Machine-achieved the highest recall of 100%. Experimental results show that, despite the high precision of complex models, clinical utility favors simple linear models that reliably identify all positive cases. The presented study presents a reliable, reproducible, transparent, and robust pipeline with strong potential for clinical diagnostic support in MPM.