Copyright © 2026 Authors retain the copyright of this article. This article 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.

@article{197022,
        author = {Sherine Sheeba Grace A and Prasanna Lakshmi M and Swetha S},
        title = {StudyFort: A Browser-Native, Hardware-Independent Framework for Continuous Cognitive Load Estimation via Multi-Modal Signal Integration},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {11},
        pages = {4740-4751},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=197022},
        abstract = {Existing solutions for cognitive load quantification in educational environments are constrained either by dependence on specialised physiological acquisition hardware or by the reactive nature of self-report instruments. StudyFort addresses this gap through a fully browser-resident, hardware-free architecture capable of estimating cognitive load continuously and with low latency. The system draws on three concurrently operating signal channels: three-dimensional facial landmark coordinates extracted at 30 FPS from a standard webcam via the TensorFlow.js FaceMesh model (468 landmarks per frame), keystroke inter-event timing patterns, and scroll-rate dynamics. A multi-dimensional Kalman filter provides minimum-variance state estimation across these channels, while Fast Fourier Transform power spectral density analysis quantifies fatigue-related spectral shifts in the blink-rate time series. Blink events are characterised through the Eye Aspect Ratio metric derived from six periocular landmarks. A Random Forest classifier, trained on labelled data collected from actual study participants, maps a nine-dimensional processed feature vector to one of three cognitive states: Focused, Relaxed, or Overloaded. The classifier achieves 85.3% accuracy and a Cohen's Kappa of 0.78 on a participant-independent held-out test partition. The system operates at 30 FPS with end-to-end latency of 50.2 ms, satisfying the design target of 200 ms. External validation via the NASA Task Load Index yields a Pearson correlation of r = 0.73 (p < 0.001) and state-level agreement of 83.3%, confirming strong correspondence between automated system outputs and subjective workload ratings across three experimental conditions.},
        keywords = {cognitive load monitoring, browser-based sensing, multi-modal signal fusion, Kalman filter, Fast Fourier Transform, eye aspect ratio, Random Forest classification, educational technology},
        month = {April},
        }