Cognitive load, emotion, and human–computer interaction (HCI) behaviors are deeply interlinked psychological and behavioral constructs that influence learning performance. Recent advances in behavioral biometrics—particularly keystroke dynamics and mouse movement analytics—have enabled non-intrusive, real-time inference of cognitive and affective states. This review paper synthesizes findings from major studies across cognitive load theory, affective computing, keystroke and mouse dynamics, educational data mining, and adaptive learning systems. We integrate results from simulation-based medical training, grammar learning, programming education, mental arithmetic tasks, and large-scale keystroke datasets to demonstrate how user interaction patterns reliably encode markers of stress, engagement, frustration, and proficiency. Building on these insights, we propose a unified framework for next-generation adaptive e-learning systems capable of continuous monitoring, state recognition, and intervention. This expanded review consolidates contributions, identifies research gaps, and outlines future directions to advance personalized education and intelligent tutoring.