Predicting Student Cognitive Overload Using Multimodal Behavioral Time-Series Modeling

Abstract

Cognitive overload is a major contributor to student stress, burnout, and disengagement, yet most educational support systems respond only after performance degradation occurs. This study investigates whether cognitive overload can be predicted in advance using temporal patterns in student planning behavior. I analyzed multimodal behavioral data derived from task management activity, including deadline density, task rescheduling frequency, completion latency, and self-reported stress indicators. Using engineered time-series features, I trained and evaluated multiple machine learning models, including logistic regression, random forest classifiers, and Long Short-Term Memory (LSTM) networks. Experimental results demonstrate that temporal models significantly outperform static baselines, with the LSTM achieving an AUROC of 0.86. Feature trend analysis reveals systematic behavioral drift in the days preceding overload events, suggesting that overload emerges gradually rather than abruptly. These findings highlight the feasibility of early-warning systems for student overload and motivate adaptive planning tools that intervene proactively rather than reactively.

Introduction

The text presents a study on early detection of student cognitive overload using behavioral data from digital task management platforms. Cognitive overload negatively impacts academic performance, decision-making, mental health, and student retention, yet current educational systems rely mainly on retrospective indicators such as grades and missed deadlines, which are ineffective for early intervention.

The study proposes that cognitive overload develops gradually and is preceded by observable changes in planning behavior, including increased task rescheduling, higher deadline density, and delayed task completion. Using anonymized data from 312 university students over one academic year, the research models these behaviors as time-series data to predict overload before performance declines.

Five key behavioral features were engineered—deadline density, rescheduling frequency, completion latency, task load variance, and stress index—and analyzed using both static models (logistic regression and random forest) and a temporal LSTM model. Results show that the LSTM significantly outperforms static models (AUROC = 0.86), confirming the importance of temporal context in capturing behavioral drift.

Analysis reveals that rescheduling frequency and deadline density are the strongest predictors of overload, with warning signs appearing 10–14 days in advance, providing a practical intervention window. The findings demonstrate that non-invasive, interpretable behavioral features, derived from routine academic activity, can effectively support early-warning systems without intrusive data collection.

The study concludes that integrating temporal models into educational platforms could enable proactive support such as workload adjustments or counseling referrals. However, limitations include reliance on observational data, partial dependence on self-reported stress, single-institution data, selection bias, and lack of real-world deployment and intervention validation. Future work is recommended to address these gaps through broader validation and experimental intervention studies.

Conclusion

This paper demonstrates that cognitive overload can be predicted using temporal patterns in student planning behavior with meaningful accuracy and sufficient advance notice to enable intervention. By framing overload as a time-series classification problem, I showed that sequential models substantially outperform static approaches, achieving strong predictive performance (AUROC = 0.86) using interpretable behavioral features derived from routine academic activities. More broadly, this work emphasizes the value of designing educational systems that observe how students behave over time rather than reacting only after failure occurs. The progressive nature of overload onset, characterized by measurable behavioral drift beginning 10-14 days before critical events, suggests a meaningful window for proactive intervention that current reactive systems fail to exploit. The finding that rescheduling frequency and deadline density serve as leading indicators of cognitive overload has important implications for educational platform design. Rather than treating task management tools as passive recording systems, platforms could actively monitor these patterns and provide adaptive support such as workload visualization, pacing recommendations, or connection to institutional resources. However, predictive models alone are not the solution. Their true value lies in enabling systems that adapt to students\' lived realities, adjusting expectations and support dynamically. Understanding how overload emerges is a necessary but insufficient step toward building tools that help students sustain productivity without sacrificing well-being. The ethical deployment of such systems requires careful consideration of privacy, autonomy, and the potential for surveillance or coercion. Future research should focus on closing the loop between prediction and intervention, developing adaptive systems that not only identify at-risk students but also provide personalized support mechanisms whose effectiveness has been empirically validated. Additionally, investigating the generalizability of these temporal patterns across diverse educational contexts and student populations remains an important priority. Ultimately, the goal is not simply to predict when students will struggle, but to create educational environments that prevent overload through better workload management, improved support structures, and systems that recognize and respond to early warning signs before a crisis occurs. This work provides a foundation for such systems by demonstrating the technical feasibility of early overload detection using passively collected behavioral data.

References

[1] S. Hochreiter and J. Schmidhuber, \"Long short-term memory,\" Neural Computation, vol. 9, no. 8, pp. 1735-1780, 1997. [2] R. S. Baker and P. S. Inventado, \"Educational data mining and learning analytics,\" in Learning Analytics: From Research to Practice, J. A. Larusson and B. White, Eds. New York: Springer, 2014, pp. 61-75. [3] R. S. Lazarus, Stress and Emotion: A New Synthesis. New York: Springer Publishing Company, 1999. [4] R. F. Kizilcec, C. Piech, and E. Schneider, \"Deconstructing disengagement: Analyzing learner subpopulations in massive open online courses,\" in Proc. 3rd Int. Conf. Learning Analytics and Knowledge, 2013, pp. 170-179. [5] S. Y. Yoo and Y. Huang, \"Comparison of deep learning models for real-time emotion recognition,\" in Proc. IEEE Int. Conf. Big Data and Smart Computing, 2018, pp. 725-728. [6] J. Huang and N. Li, \"Predicting student performance using multi-modal physiological signals,\" IEEE Trans. Learning Technologies, vol. 12, no. 3, pp. 378-389, July 2019. [7] A. Bosch and F. D\'Mello, \"The affective experience of novice computer programmers,\" Int. J. Artificial Intelligence in Education, vol. 27, no. 1, pp. 181-206, 2017. [8] C. Romero and S. Ventura, \"Educational data mining: A survey from 1995 to 2005,\" Expert Systems with Applications, vol. 33, no. 1, pp. 135-146, 2007. [9] D. Gaševi?, S. Dawson, and G. Siemens, \"Let\'s not forget: Learning analytics are about learning,\" TechTrends, vol. 59, no. 1, pp. 64-71, 2015. [10] Y. LeCun, Y. Bengio, and G. Hinton, \"Deep learning,\" Nature, vol. 521, pp. 436-444, May 2015. [11] N. V. Chawla, K. W. Bowyer, L. O. Hall, and W. P. Kegelmeyer, \"SMOTE: Synthetic minority over-sampling technique,\" J. Artificial Intelligence Research, vol. 16, pp. 321-357, 2002. [12] J. Sweller, \"Cognitive load theory, learning difficulty, and instructional design,\" Learning and Instruction, vol. 4, no. 4, pp. 295-312, 1994.

Copyright

Copyright © 2026 Nidhish Somi Umanath Durgadevi. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.