Performance Analysis of ML Models: An Overlook through the Perspective of VOCs to Detect Lung and Liver Cancer

This study highlights the role of exhaled breath analysis as one promising, non-invasive technique used in the detection of cancer biomarkers, specifically focusing on Volatile Organic Compounds (VOCs) linked to lung and liver cancers. Three machine learning models were incorporated for model testing- Random Forest, Decision Tree, and SVM to identify the best classification models among these that distinguish the cancer profiles. The study achieved the highest accuracy of 95% for liver cancer and accuracy of rate 85% for lung cancer. Multivariate heatmaps and robust correlation matrices were plotted to highlight the unique VOC patterns which distinguish between benign, cancerous and healthy profiles. Advanced data imputation techniques were carefully evaluated, with zero imputation emerging as an effective method for optimizing data dimensionality. The VOCs associated with lung and liver carcinomas were condensed from 27 to 20 and 16 to 10, respectively, with compounds being shown as the most diagnostically significant. These findings support VOC profiling as a novel, innovative and patient-friendly approach for early cancer detection that could transform clinical oncology practices.