Predictive Analytics in Healthcare: A Hybrid Model for Efficient Medical Cost Estimation for Personalized Insurance Pricing

Abstract Predictive analytics in healthcare enables personalized insurance pricing by analyzing diverse datasets, including patient histories and treatment outcomes. In the recent years, Machine Learning (ML) models and statistical approaches are used to enhance predictive analytics in healthcare by accurately identifying individual risk factors and forecasting medical costs from complex datasets, enabling more personalized insurance pricing. However, these models are not adept in capturing the intricate relationships between linear and non-linear patterns within healthcare data. This limitation can restrict their effectiveness in personalizing insurance pricing to the fullest extent, underscoring the need for innovative solutions that can bridge this gap. This paper presents the Linear Regression-Gaussian Deep Belief Network (LR-GDBN), a hybrid predictive model developed to improve healthcare analytics by accurately forecasting medical costs. This model effectively captures linear and non-linear patterns in healthcare data, surpassing traditional methods. LR-GDBN is evaluated with a standard dataset, and it demonstrates a low prediction error of 0.4926 surpassing conventional statistical and machine learning models. Unlike the state-of-the-art models capable of handling only non-linear health data in insurance pricing, this model featuring ability to handle linear and non-linear data is unique of its kind in the domain of healthcare underwriting.