Performance comparison of feature selection algorithms in the life expectancy risk prediction of post-operative lung cancer patients

Misra A.; Torres-Cruz F.; Pedro R.; Laura-Murillo; Pari-Condori E.Y.; Dixit C.K.; Puma J.C.T. · AIP Conference Proceedings (2023) · DOI: 10.1063/5.0126447

Resumen

Prediction of mortality risk for post-operative lung resection patients with scores may recommend the physicians to change the course of treatment to increase the life expectancy. Numerous post-operative scores such as biochemical, demographic and spirometric variables were used to predict the mortality risk and treatment post-surgery. Such large number of variables misleads the treatment as well as increase the complexity of decision making processes. In this paper, the prediction accuracy and complexity of support vector machine classification model before and after feature ranking was studied. Four feature ranking algorithms such as oneR, gain ratio, information gain and symmetrical uncertainty were used to rank the feature subsets. The classification accuracy and complexity before the feature selection was 84.89% and 78,264 kernels evaluated, respectively. However, after feature selection, the accuracy improved to 85.14% and complexity significantly reduced to ~40% due to the reduction in number of kernels evaluated. These results highlighted the importance of feature selection to predict the life expectancy of post-operative lung cancer patients.

Instituciones: Universidad Nacional del Altiplano