Reinforcing synthetic data for meticulous survival prediction of patients suffering from left ventricular systolic dysfunction

Congestive heart failure is among leading genesis of concern that requires an immediate medical attention. Among various cardiac disorders, left ventricular systolic dysfunction is one of the well known cardiovascular disease which causes sudden congestive heart failure. The irregular functioning of a heart can be diagnosed through some of the clinical attributes, such as ejection fraction, serum creatinine etcetera. However, due to availability of a limited data related to the death events of patients suffering from left ventricular systolic dysfunction, a critical level of thresholds of clinical attributes can not be estimated with higher precision. Hence, this paper proposes a novel pseudo reinforcement learning algorithm which overcomes a problem of majority class skewness in a limited dataset by appending a synthetic dataset across minority data space. The proposed pseudo agent in the algorithm continuously senses the state of the dataset (pseudo environment) and takes an appropriate action to populate the dataset resulting into higher reward. In addition, the paper also investigates the role of statistically significant clinical attributes such as age, ejection fraction, serum creatinine etc., which tends to efficiently predict the association of death events of the patients suffering from left ventricular systolic dysfunction