Perceived AI Feedback Quality, Learner Trust in AI, and Self-Regulated Learning as Predictors of Mathematics Learning Outcomes

Abstract

This study examined how Perceived AI Feedback Quality (PAQ), Learner Trust in AI (LAT), and Self-Regulated Learning (SRL) influence Mathematics Learning Outcomes (MLO) among undergraduate mathematics students in Ghana. A quantitative cross-sectional survey design was used. Using a structured online questionnaire, data were collected from 298 undergraduate mathematics students at AAMUSTED, selected from a population of 1,202 students. PAQ, LAT, SRL, and MLO were measured with validated Likert-type scales adapted from prior studies. EFA supported a four-factor structure, with KMO = 0.929, a significant Bartlett’s test (p < .001), and 76.08% cumulative variance explained. CFA indicated excellent model fit (CMIN/df = 1.19, RMSEA = 0.025, SRMR = 0.0296, CFI = 0.993, GFI = 0.949, TLI = 0.992). All constructs exhibited strong reliability (Cronbach’s α = 0.898–0.916; CR = 0.90–0.92) and convergent validity (AVE = 0.67–0.72), while HTMT values below 0.85 confirmed discriminant validity. SEM results showed that PAQ had a non-significant direct effect on MLO, but significantly predicted LAT and SRL. Both LAT and SRL significantly and positively predicted MLO. Mediation analysis using 99% bias-corrected bootstrapped confidence intervals revealed that PAQ influenced MLO indirectly through SRL, and that PAQ also enhanced SRL indirectly via LAT. These findings suggest that high-quality AI-generated feedback improves mathematics achievement primarily by building learners’ trust in AI and strengthening their self-regulated learning, rather than through a direct path from feedback quality to performance. The study underscores the need for AI feedback tools and instructional practices in African higher education that explicitly foster trust and self-regulation.