Digital technologies have transformed how information is accessed and shared; however, individuals with disabilities continue to face significant barriers in engaging with educational content. The shift to electronic formats presents an opportunity to promote inclusion by integrating accessibility features. Accessible e-books are essential for fostering equitable participation in learning environments, supporting lifelong learning, and aligning with global sustainability goals such as SDG 4 (Quality Education) and SDG 10 (Reduced Inequalities). While international frameworks like the Web Content Accessibility Guidelines (WCAG) advocate for accessible publishing, significant gaps remain. This paper presents a Systematic Literature Review (SLR) aimed at examining the role of accessibility features in shaping the quality, usability, and affordability of e-books. Based on well-defined guidelines, we defined and executed an SLR protocol, identifying 1618 studies in seven different scientific databases. At the end of the review process, we selected 74 studies to answer four research questions related to the accessibility features employed by e-books and their reading platforms. By synthesizing existing evidence, this review highlights barriers, best practices, and strategic pathways to inform educators, publishers, and policymakers. Ultimately, this study advocates for the development of inclusive and sustainable digital content ecosystems that meet the diverse needs of all learners.

The rapid integration of digital technologies into education has transformed electronic textbooks (e-textbooks) from static replicas of printed books into interactive, multimedia-rich learning environments. This bibliometric literature review synthesizes findings from 89 peer-reviewed studies published between 2000 and 2024, retrieved from IEEE Xplore, ACM Digital Library, ScienceDirect, and SpringerLink. Following the Kitchenham and Charters methodology and complemented by bibliometric analyses, this review is guided by the overarching research question (RQ): How are e-textbooks conceptualized, designed, and deployed to support learning outcomes while promoting accessibility and equity? This guiding question is complemented by three additional RQs: (1) How has the field of e-textbooks developed over time? (2) Which countries are leading research related to interactive textbooks, considering quantitative indicators of publications/country? (3) What evidence exists regarding the contribution of e-textbooks to learning outcomes? The analysis reveals a clear trajectory from early static digitization toward intelligent platforms incorporating adaptive feedback, personalization, and analytics. Well-designed e-textbooks using multimedia learning principles demonstrate significant potential to enhance conceptual understanding, learner engagement, and motivation—especially in K–12 contexts. However, the field remains challenged by definitional inconsistencies, the uneven application of accessibility standards, and the geographic concentration of research output in North America and East Asia. These disparities highlight an urgent need to align digital textbook development with broader global education goals. By foregrounding accessibility, inclusion, and equitable access, e-textbooks can play a transformative role in advancing Sustainable Development Goal 4 (Quality Education) by ensuring inclusive, engaging learning opportunities, and SDG 10 (Reduced Inequalities) by reducing educational disparities across diverse learner populations. This review concludes by calling for interdisciplinary frameworks that integrate cognitive science, human–computer interaction, and artificial intelligence to guide the design of the next generation of interactive, accessible, and pedagogically robust e-textbooks—tools that are not only technologically innovative, but also aligned with global commitments to equitable and sustainable education for all.

Ensuring the availability of didactic materials within Brazil’s National Textbook Program (PNLD) depends on reliable school enrollment forecasts. However, administrative time series often exhibit limited historical coverage and heterogeneous stability, which can constrain predictive accuracy. This study investigates the impact of data quality on forecasting performance using Brazilian School Census enrollment records from 2007 to 2023. We stratify school–stage series into three data-quality regimes defined from the available records, based on historical availability (number of observed years) and stability (coefficient of variation (CV)): Light (at least 4 years; CV up to 1.0), Medium (at least 5 years; CV up to 0.5), and Rigorous (at least 8 years; CV up to 0.3). We adopt a one-step-ahead protocol ( $h=1$ ) evaluated on two horizons: a pre-pandemic forecast targeting 2019 (training through 2018) and a post-onset forecast targeting 2020 (training through 2019), the latter potentially affected by the COVID-19 pandemic in Brazil. We evaluate 14 predictive models spanning classical statistical methods, machine learning algorithms, and fuzzy time series (FTS) architectures, using mean absolute error (MAE) as the primary metric. Results show that regime-driven effects are consistent across both horizons: moving from the Light to the Rigorous regime reduces MAE by approximately 30% for 2020 and 32% for 2019, while the best-to-worst model spread within a single regime is at most 4.6 points. This gain, however, comes at a substantial coverage cost—the eligible dataset shrinks from 77,315 to 8677 series (11.2%). Under the highest-quality regime, FTS models show robust performance, with SeasonalFTS achieving the best MAE across both horizons (13.00 for 2019; 12.39 for 2020). To our knowledge, this is the first large-scale empirical quantification of the coverage–precision trade-off induced by data-quality stratification in administrative enrollment forecasting. The findings support quality-aware forecasting strategies for PNLD planning.

Efficient resource allocation in public education intrinsically depends on accurate enrollment demand forecasting. This work proposes a Data Science approach to forecast enrollments in Brazilian basic education, using microdata from the School Census from 2011 to 2021. GPU-accelerated Machine Learning techniques were employed, specifically Gradient Boosting (XGBoost and LightGBM) combined via Voting Ensemble with hyperparameter optimization via Optuna. The study applied rigorous temporal cross-validation (Walk-Forward Validation) to minimize data-leakage problems, utilizing context-specific feature engineering, including student–teacher ratios and demand acceleration. The results demonstrated high precision in regular teaching stages ( $R^2>0.97$ ) but revealed high hidden volatility in vocational education (IN_PROF_TEC: $R^2=0.528\pm 0.171$ ) when subjected to honest temporal validation. Interpretation via SHAP values indicated that historical inertia dominates in stable grades (e.g., lags represent 65% of importance in IN_FUND), while exogenous factors predominate in elective series, offering public managers a transparent tool to differentiate between structural demands and conjunctural fluctuations.