Investigating AI-Driven Adaptive Interfaces to Enhance User Engagement

Abstract

This paper presents a large-scale experimental investigation into how AI-driven adaptive user interfaces can improve user engagement in data-intensive tasks. Over 60 participants engaged with both static and dynamically adaptive dashboards; the latter reordered widgets in real time using a LightGBM model trained on preliminary interaction logs. We collected objective metrics (task completion time, error rates) and subjective measures (NASA-TLX, User Engagement Scale). The adaptive condition exhibited a 14 % reduction in completion time and a 22 % increase in reported engagement (p ¡ 0.01), alongside a 17 % decrease in perceived workload. Qualitative feedback highlighted enhanced perceived intuitiveness and satisfaction. We situate our contributions within emerging AI-HCI research, extend theory on personalization reflexives, and propose guidelines for deploying real-time adaptation responsibly. Our findings confirm that machine-learning-powered adaptation is both feasible and beneficial for enriching user experience in professional analytics tools. We conclude by discussing implications for explainability, privacy, and longitudinal viability.