Abstract
Virtual Learning Environments now record behavioral data at a scale traditional classrooms never reached, yet most online platforms still serve every student the same content, pacing, and evaluation. The learners who most need adaptive support pay the highest price for this uniformity. In the Open University Learning Analytics Dataset (OULAD), students who self-declare a disability pass 10.0 percentage points less often than their non-disabled peers and withdraw 9.1 points more often. We present an Internet of Behaviors (IoB) analytics ecosystem for adaptive e-learning that turns this behavioral record into timely, accountable support, closing the loop between observation and intervention. The ecosystem links four working parts: a forecasting model that predicts drop-out, pass/fail, and near-term disengagement from routine course activity; an explanation layer that makes each prediction legible to the instructor who has to act on it; a fairness audit centered on students with a declared disability; and a tamper-evident record of every automated decision, so that actions affecting a learner's path can be reviewed under GDPR Article 22 and FERPA. We evaluate the analytical layers on OULAD using multi-seed runs with confidence intervals and cross-validation across unseen course presentations. The model holds up on those presentations (drop-out AUC 0.916), and its explanations agree closely across methods. The fairness gaps it leaves stay small. Three results complicate the story: a strong gradient-boosting baseline outperforms the deep model on drop-out; an external sentiment signal adds nothing over a simple week indicator; and the multi-task design is smaller and faster rather than more accurate. On a public blockchain testnet, each decision was anchored within a single block at a median 23,965 gas. Inclusion and accountability can be designed into an e-learning analytics pipeline and then measured.