Explainable AI in Educational Assessment: Building Student Trust in AI-Driven Feedback through Design-Based Research

AI-driven assessment systems increasingly mediate feedback in higher education, yet the opacity of algorithmic decision-making raises concerns about student trust and the pedagogical value of automated feedback. This paper reports a design-based research study that developed and iteratively evaluated an explainable AI (XAI) feedback module for an examination preparation system at Akanu Ibiam Federal Polytechnic, Unwana, Nigeria. Across two design cycles involving 36 students, we tested four XAI principles: diagnostic explanation, contrastive explanation, calibrated confidence, and human corroboration. Data were collected through think-aloud protocols, semi-structured interviews, system logs, pre/post topic tests, and a trust scale. Findings indicate that diagnostic and contrastive explanations significantly improved students' comprehension of errors and their willingness to act on feedback. Calibrated confidence displays reduced over-trust, while instructor annotations were valued but inconsistently accessed. Cycle 2 refinements, including simplified language and visual cues, improved usability ratings and feedback uptake. The study contributes an empirically grounded framework for pedagogically meaningful XAI in resource-constrained higher education contexts and demonstrates the value of design-based research for developing context-sensitive AI assessment tools.