Date of Publication : 29th May 2025

Abstract: Hyperforecasting allows medical professionals and individuals to make decisions about health care, resources, and planning. Yet, actuarial and statistical science methods are generally ineffective in considering various unique, unexpected, and nonlinear components of longevity. New ML algorithms have appeared lately, permitting life expectancy estimates that can adjust to each individual and change as new information comes in. This review tested how Random Forest, Extreme Gradient Boosting, and Neural Networks could forecast each individual's life expectancy in various clinical and demographic settings. The review collects results from peer-reviewed papers published from 2015 to 2025 in PubMed, IEEE Xplore, and Scopus databases. Accuracy, adaptability, how well the machine can be interpreted, and whether it can be easily integrated are measured. Random forests are respected for being sturdy and readable, which is not something XGBoost is known for, but they seem more accurate and performant when using complex data. The greatest potential of deep learning networks is modeling difficult data, but they are less transparent and demand lots of information to function properly. ML models are more accurate than many older techniques, yet there are problems with explaining them, ensuring their ethics, and being fair. By reviewing these findings, we can recognize that it's important for AI to be clinically understandable and highlight designs that may improve them in the future.

Reference :