Feature selection for classification of survival analysis in lymphoma cancer

Survival analysis is known to have great potential for classifying clinical dataset and it has not been fully explored for classifying lymphoma cancer. Most survival analysis methods such as Life Table and Kaplan-Meier estimator have the problem of producing survival prognosis because they can only estimate the probability of survival and are limited due to their dependence on a type of lymphoma features. Thus, this research applied Particle Swarm Optimization (PSO) feature selection for Support Vector Machine (SVM) classification to address the estimation problem and limitation of survival analysis of lymphoma cancer. The PSO-SVM is a procedure that identifies the most significant lymphoma information in terms of lymphoma features by calculating the fitness values and accuracy average of the features classification. Based on the values obtained, then the most significant lymphoma features that increased the performance of survival classification in terms of accuracy would be identified. In this research, PSO-SVM outperformed classification of SVM without feature selection with improvements of survival accuracy between 0.09% and 22.28%. Apart from that, the survival analysis results could be extracted and interpreted into a more understandable explanation by using Decision Tree. These findings have shown that the proposed PSO-SVM is capable of reducing the number of lymphoma features by obtaining the most accurate classification of survival analysis as compared to classification that uses all the lymphoma feature