Artificial Intelligence–Augmented Statistical Inference: Integrating Machine Learning with Classical Estimation and Hypothesis Testing

Abstract

The convergence of machine learning and classical statistical inference has given rise to a new paradigm in data analysis, enabling researchers to address high-dimensional, complex, and non-linear data structures that challenge traditional inferential techniques. While classical methods such as maximum likelihood estimation and hypothesis testing provide strong theoretical guarantees and interpretability, they often struggle in modern data-rich environments. Conversely, machine learning models offer remarkable predictive power but frequently lack transparency and rigorous uncertainty quantification. This study examines artificial intelligence–augmented statistical inference as a principled integration of these two approaches, aiming to preserve statistical rigor while enhancing flexibility and performance. By synthesizing theoretical foundations and empirical evidence across diverse application domains—including neuroimaging, model-based deep learning systems, and survival analysis—the paper demonstrates how hybrid frameworks can improve estimation accuracy, hypothesis testing robustness, and interpretability. The analysis highlights the role of model-based deep learning, penalized estimation, and machine learning–assisted feature selection in maintaining inferential validity and uncertainty quantification. Overall, this work establishes AI-augmented statistical inference as a robust and interpretable framework for advancing modern data-driven science while retaining the foundational principles of classical statistics.