A Comparative Study of Explainable AI Techniques for Healthcare Predictive Analytics

Clinical decision-making increasingly relies on predictive models, yet these tools often lack adequate explanation mechanisms and are thus ineligible for real-world adoption. Restrictions on model temporality, population coverage, and data privacy inhibit their use during training; only post-training use of fixed, less sensitive information remains feasible. To counsel such models, the adoption of Explainable AI (XAI) techniques emerges as the leading recourse and has already generated considerable interest. Here, a systematic overview of XAI techniques for the tabular datasets characteristic of healthcare predictive analytics is provided. Because these datasets differ significantly from other data modalities in structure, content, and intended use, XAI approaches developed for computer vision analyses, natural language processing, and other domains are not well suited to healthcare needs. The inclusion of past patient states, health trajectory histories, and other types of temporal data further differentiates healthcare from typical settings such as bank fraud detection. Formulating an appropriate, comprehensive classification of XAI methods for predictive analytics in general and healthcare in particular-therefore constitutes a central challenge.