PERFORMANCE OF RANDOM FORESTS, EXTREME GRADIENT BOOSTING AND SUPPORT VECTOR MACHINES EMPLOYED IN LIPIDOMICS

AUTHOR(S) / АУТОР(И): David Pirić and Romana Masnikosa

Download Full Pdf   

DOI: 10.46793/Phys.Chem24I.223P

ABSTRACT / САЖЕТАК:

Herein we present the performance of three supervised machine learning (ML) algorithms: random forests (RF), extreme gradient boosting (XGB) and support vector machines (SVM) in classification of human serum samples into pancreatic cancer or control group, using a lipidomic dataset retrieved from the research article „Lipidomic profiling of human serum enables detection of pancreatic cancer“ by Wolrab et al. [1]. Our main objective was to assess and compare, for the three ML techniques, the performance metrics, that is accuracy, precision, sensitivity, F1 score and ROC- AUC, with those computed by decision trees (DT), which is the basis of RF and XGB algorithms, and commonly used partial least squares – discriminant analysis (PLS-DA) and orthogonal projections to latent structures – discriminant analysis (OPLS-DA). We suggest that RF, XGB and SVM represent excellent binary classifiers, making these three promising candidates for future use in the discovery of potential lipid biomarkers.

KEYWORDS / КЉУЧНЕ РЕЧИ:

ACKNOWLEDGEMENT / ПРОЈЕКАТ:

This research is funded by the Ministry of Science, Technological Development and Innovation, Republic of Serbia, Grants: No. 451-03-66/2024-03/200017.

REFERENCES / ЛИТЕРАТУРА:

• D. Wolrab et al., Lipidomic profiling of human serum enables detection of pancreatic cancer, Nature Communications, 13 (2022), 124