Adverse selection remains a significant challenge in the insurance industry, often resulting in substantial financial losses for insurers. The primary hurdle in addressing the issue lies in accurately identifying and quantifying adverse selection. Traditional methods often fail to adequately account for the heterogeneity of insurance purchasers and the endogenous nature of their insurance decisions. This study introduces an innovative approach that integrates the Gaussian Mixture Model and the regression-based model from Dionne et al. (2001) to assess adverse selection, addressing the limitations of previous methods. Through comprehensive simulations, we demonstrate that our method yields unbiased estimates, outperforming existing approaches. Applied to China’s automobile insurance market, leveraging IoT devices to track telematics data, this method captures risk heterogeneity among the insured. The results offer robust evidence of adverse selection, in contrast to conventional methods that fail to detect this phenomenon due to their inability to capture the underlying relationship between customer risk and claim behavior. Our approach offers insurers a robust framework for identifying information asymmetries in the market, thereby enabling the development of more targeted policy interventions and risk management strategies.