Adaptive thresholding for anomaly detection in highdimensional claim datasets

Anomaly detection in high-dimensional healthcare and insurance claim datasets presents significant challenges due to data sparsity, heterogeneity, and the presence of complex, nonlinear relationships. Traditional fixed-threshold methods often fail to capture subtle deviations, leading to high false-positive or false-negative rates. This study explores adaptive thresholding as a dynamic approach for identifying anomalous claims in high-dimensional spaces. By leveraging statistical distribution analysis, local density estimation, and machine learning–driven calibration, adaptive thresholds are tuned in response to data characteristics rather than predefined static rules. Experiments on simulated and real-world claim datasets demonstrate that adaptive thresholding improves detection accuracy, scalability, and robustness compared to conventional methods. The findings highlight its potential for enhancing fraud detection, reducing operational inefficiencies, and supporting data-driven decision-making in large-scale insurance systems