Survival Prediction of Cervical Cancer Patients using Genetic Algorithm-Based Data Value Metric and Recurrent Neural Network

Abstract

Survival analysis and machine learning has been shown to be an indispensable aspect of disease management as it enables practitioners to understand and prioritize treatment mostly in terminal diseases. Cervical cancer is the most common malignant tumor of the female reproductive organ worldwide. Survival analysis which is a time –to –event analysis for survival prediction is therefore needed for cervical cancer patients. Data Value Metric (DVM) is an information theoretic measure which uses the concept of mutual information and has shown to be a good metric for quantifying the quality and utility of data as well as feature selection. This study proposed the hybrid of Genetic Algorithm and Data Value Metric for feature selection while Recurrent Neural Network and Cox Proportionality Hazard ratio was used to build the survival prediction model in managing cervical cancer patients. Dataset of 107 patients of cervical cancer patients were collected from University of Benin Teaching Hospital, Benin, Edo State and was used in building the proposed model (RNN+GA-DVM). The proposed system outperform the existing system as the existing system had accuracy of 70% and ROC score of 0.6041 while the proposed model gave an accuracy of 75.16% and ROC score of 0.7120 respectively. From this study, It was observed using the GA_DVM features selection that the variables highly associated with cervical cancer mortality are age_at_diagnosis, Chemotherapy, Chemoradiation, Histology, Comorbidity, Menopause, and MENO_Post. Thus, with early diagnosis and proper health management of cervical cancer, the age of survival of cervical cancer patients can be prolonged.