Can learning analytics actually improve student outcomes?
Yes, and the effect can be substantial. A meta-analysis of 24 randomized studies found that AI chatbots had a large positive effect on students' learning outcomes overall [2]. The effect was especially strong in higher education and for short interventions, suggesting that novelty and engagement play a role [2]. Similarly, a systematic review of mobile learning environments from 2014 to 2023 reported moderate to significant positive effects across different educational levels and subjects [4]. Personalized learning, powered by analytics, also shows promise: a review of 144 articles found that the most common goals of learning analytics are enhancing learning experience and providing personal recommendations [11].
However, the evidence is not uniform. While many studies report gains, the same review noted that predictions from analytics are often not clearly translated into pedagogical actions [1]. This means the potential is real, but realizing it requires more than just collecting data—it requires actionable insights for teachers and students.
Does learning analytics risk widening inequality?
Yes, the risk is well-documented. A 2025 study using the Open University Learning Analytics Dataset found that a Random Forest model predicting student outcomes achieved 90.2% accuracy and 93.3% recall, but fairness analysis revealed a disparate impact ratio of just 0.25 [3]. This means students without prior qualifications and from lower socioeconomic deciles were four times less likely to be correctly identified as passing compared to their peers—a clear bias [3]. Another study on AI in Indian education noted that only about 57% of schools have computers and 54% have internet connectivity, meaning digital tools can easily exclude the most disadvantaged students [5].
The problem is not just about access. Even when data is available, algorithms can encode existing biases. A fairness-aware study on math learning outcomes found that standard logistic regression, support vector machines, and random forests all produced biased predictions across demographic subgroups [9]. The authors concluded that without explicit fairness constraints, AI can increase educational inequality [9]. This is echoed in a review of AI in school education, which highlighted algorithmic bias as a critical concern [6].
Can we have both improvement and equity?
Yes, but it requires deliberate design and often involves trade-offs. The same study that found a disparate impact ratio of 0.25 tested two mitigation strategies: pre-processing reweighing preserved accuracy while modestly reducing error rate disparities, and post-processing thresholding improved the disparate impact ratio to 0.45—a meaningful gain—but at the cost of reducing recall from 93.3% to a lower value [3]. This illustrates a fundamental tension: you can make predictions fairer, but you may lose some sensitivity in identifying at-risk students.
Other approaches show more promise. A study on fair logistic regression (Fair-LR) demonstrated that it could achieve both desirable predictive accuracy and better fairness compared to standard models [9]. Similarly, an equity-forward learning analytics dashboard was designed specifically to support marginalized students, revealing previously hidden inequities in course outcomes and enabling targeted interventions [8]. The key is to embed fairness principles from the start, not as an afterthought. Ethical frameworks for learning analytics emphasize data privacy, informed consent, transparency, and fairness as core principles to prevent the reinforcement of existing inequalities [7].
The evidence also points to the importance of context. In India, AI-driven reforms under the National Education Policy 2020 aim to personalize learning and bridge linguistic barriers, but the digital divide remains a major obstacle [5][10]. In Australia, a survey of over 151,000 graduates found that access to work-integrated learning—a form of experiential education—is not uniform across student groups, and tailored approaches are needed to ensure equitable outcomes [12]. Across all these contexts, the message is consistent: learning analytics can improve outcomes, but only if equity is an explicit design goal, not an assumed side effect.
About These Sources
This answer is built on 12 peer-reviewed studies — published from 2022 to 2026, 8 from 2024 or later, 4 in Q1 journals, collectively cited 537 times — selected as the most relevant from 15 studies that passed quality screening, drawn from 66 papers retrieved from a database of over 500 million.
Sources used in this answer
A Current Overview of the Use of Learning Analytics Dashboards
A 2024 overview of systematic reviews on learning analytics dashboards found that while research is growing rapidly, predictions are often not translated into pedagogical actions, and issues of inequality, data ownership, and privacy remain unresolved [1].
Do <scp>AI</scp> chatbots improve students learning outcomes? Evidence from a meta‐analysis
A meta-analysis of 24 randomized studies found that AI chatbots had a large positive effect on learning outcomes, with stronger effects in higher education and for short interventions [2].
Fairness-Aware AI in Education: Detecting and Reducing Bias in Student Assessment Systems
A 2025 study using the Open University Learning Analytics Dataset found that a Random Forest model achieved 90.2% accuracy but had a disparate impact ratio of 0.25, meaning lower-socioeconomic students were four times less likely to be correctly predicted; post-processing improved the ratio to 0.45 but reduced recall [3].
Mobile learning evolution: a decade of developments (2014-2023)
A systematic review of mobile learning from 2014 to 2023 found moderate to significant positive effects on student outcomes, with project-based, collaborative, and situated learning enhancing effectiveness [4].
AI-driven Reforms in Indian Education Policy and Governance (2016–2026): A Review
A review of AI-driven reforms in Indian education (2016–2026) noted that only 57% of schools have computers and 54% have internet, and while AI tools show promise for personalization, the digital divide risks increasing inequity [5].
AI and school education
A review of AI in Indian school education highlighted barriers such as digital inequality, lack of infrastructure, need for teacher training, and ethical concerns including algorithmic bias and data privacy [8].
Designing Ethical Learning Analytics Frameworks to Support Decision Making and Equity in Technology Enhanced Higher Education Environments
A 2025 study proposed an ethical framework for learning analytics based on four principles—data privacy, informed consent, transparency, and fairness—to prevent reinforcement of existing inequalities [9].
Equity-Forward Learning Analytics: Designing a Dashboard to Support Marginalized Student Success
A 2024 study on an equity-forward Course Diversity Dashboard revealed previously hidden learner inequities in all courses studied, and validated the effectiveness of evidence-informed study strategies for marginalized students [10].
Using fair AI to predict students’ math learning outcomes in an online platform
A 2022 study developed a fair logistic regression (Fair-LR) algorithm that achieved both desirable predictive accuracy and better fairness compared to standard models, reducing bias in predicting math learning outcomes [11].
Artificial Intelligence in Education: Opportunities, Challenges, and Ethical Implications for Teaching and Learning in India
A 2026 review of AI in Indian education found that while AI tools can personalize learning and improve engagement, challenges include digital inequality, algorithmic bias, and the need for ethical frameworks and inclusive policies [12].
An analysis of learning analytics in personalised learning
A 2022 analysis of 144 articles on learning analytics for personalized learning found that the most common goals were enhancing learning experience and providing personal recommendations, with methods including statistical tests, classification, and clustering [13].
Equity and inclusion in work-integrated learning: participation and outcomes for diverse student groups
A survey of over 151,000 Australian graduates found that access to work-integrated learning is not uniform across student groups, and tailored approaches are needed to ensure equitable outcomes [14].