Deep Learning for Fraud Prediction in Preauthorization for Health Insurance

Abstract

Health insurance fraud remains a global menace despite the controls implemented to address it; one of such controls is preauthorization. Although, preauthorization promises reduction in fraud, waste and abuse in healthcare, it places undue administrative burden on healthcare service providers and delay in patient care. This limitation has not been thoroughly explored by works of literature in the machine learning domain. In this work, a deep learning model is proposed to learn the preauthorization process for fraud prevention in health insurance for improved process efficacy. In detail, a de-identified HMO preauthorization dataset is used for training the Long Short- Term Memory (LSTM) network. To address class imbalance and avoid data overfitting, the proposed approach utilizes random oversampling and dropout techniques respectively. The experimental results reveal that the proposed model can effectively learn preauthorization request patterns while offering a fraud detection accuracy rate of over 90% with a 2-4% improvement rate in accuracy when compared with previous techniques based on conventional machine learning techniques. The proposed technique is capable of detecting anomalous preauthorization requests based on medical necessity.