An Explainable Two-Tier Machine Learning Framework for Early Detection and Refined Assessment of Anxiety Levels

Anxiety disorders are included in the list of the most prevalent mental-health disorders in the world, but most of the time they go undetected as a diagnosis is mostly made on subjective basis and there are no widely applicable screening instruments. The study proposes a two-step, explainable machine-learning pipeline that aims at identifying anxiety at its initial stages and narrow down the process of assessing its level through a set of behavioral and psychological indicators. In the first stage, logistic regression is applied to select the participants, and the results include probability scores and the coefficients are clear, which indicate the relative weights of the key predictors of Well-Being Score, Anxiety Score, and a composite Risk Index. The cases that lie close to the decision boundary are sent to the second stage where a decision-tree classifier is used to to re-classify these borderline observations to enhance the overall accuracy. On a real-world dataset of 4,000 samples, the hierarchical structure yielded the highest accuracy of 0.92, F1-score of 0.90 and an area under the ROC curve (AUC) value of 0.94, which is higher than the conventional single-stage classifiers. The approach provides a reliable scalable basis to support mental-health screening systems that use AI to aid the practice of early intervention and personalized care by providing a combination of diagnostic reliability and interpretable, clear decision rules.