Saturation in modified correlation metrics: a reanalysis of advanced Matthews coefficient result for EEG stress classification
DOI:
https://doi.org/10.15282/isse.1.2.2026.15568Keywords:
Electroencephalography, Stress classification, Matthews Correlation Coefficient, Multimodal fusion, Power spectral densityAbstract
Mental stress is increasingly assessed through electroencephalography (EEG), which offers objective, temporally resolved measurement. The Advanced Matthews Correlation Coefficient (MCCA) has been proposed for this application, replacing the false-positive and false-negative terms of the conventional Matthews correlation coefficient with ratios of EEG arousal and valence indices raised to a fixed power, arousal classification rising from about 10% to above 98% has been reported. This study pursues two objectives: to determine analytically how MCCA behaves as its power exponent increases, and to establish when the previously reported accuracies may be generalised. The stress features examined are power spectral density values over the delta, theta, alpha and beta sub-bands of ten frontal and temporal channels, from which the arousal ratio (alpha+beta)/(delta+theta) and its left-right difference are formed. Random Forest, support vector machine, and k-nearest neighbours classifiers were evaluated under 10-fold cross-validation. MCCA is shown to converge to one of three fixed values, +1, –1, or 0, according to whether the substituted ratios exceed unity, with the true-positive and true-negative counts eliminated in the limit. The exponent needed to reach this regime scales with confusion-matrix size, so no fixed value is universally appropriate, a selection rule is derived and verified numerically. Because the limiting value is fixed by signal-derived arousal and valence ratios rather than by classifier outcomes, MCCA is better interpreted as an index of the recorded EEG signal than as a measure of classification quality. An enhanced multimodal fusion formulation is specified for future validation.
Downloads
References
[1] S. Acharya, A. Khosravi, D. Creighton, R. Alizadehsani, and U. R. Acharya, “Neurostressology: A systematic review of EEG-based automated mental stress perspectives,” Inf. Fusion, vol. 124, p. 103368, Dec. 2025, doi: 10.1016/j.inffus.2025.103368.
[2] A. Apicella, P. Arpaia, M. Frosolone, G. Masi, and E. Pasero, “Wearable electroencephalography and multi-modal mental state classification: A systematic literature review,” Comput. Biol. Med., vol. 150, p. 106134, Nov. 2022, doi: 10.1016/j.compbiomed.2022.106134.
[3] World Health Organization, Mental Health at Work: Policy Brief, Geneva, Switzerland: World Health Organization, 2022.
[4] A. M. Algarni, M. A. Aslam, M. U. Khan, A. R. Baig, and S. A. M. Al-Sarem, “Enhanced classification of video-evoked stress response using power spectral density features,” Appl. Sci., vol. 14, no. 20, p. 9527, Oct. 2024, doi: 10.3390/app14209527.
[5] F. Pinto, M. Cunha, and J. P. S. Cunha, “Stress monitoring using low-cost electroencephalogram devices: A systematic literature review,” Mar. 2024, arXiv:2403.05577, doi: 10.48550/arXiv.2403.05577.
[6] M. A. Al-Fahoum and A. A. Al-Fraihat, “Trends in EEG signal feature extraction applications,” Front. Artif. Intell., vol. 5, p. 1072801, Jan. 2023, doi: 10.3389/frai.2022.1072801.
[7] L. Zapata-Salazar, C. A. Ferrer-Riesgo, and A. A. Torres-García, “Evaluation of EEG signals by spectral peak methods and statistical correlation for mental state discrimination,” Sensors, vol. 24, no. 11, p. 3316, May 2024, doi: 10.3390/s24113316.
[8] A. Khan, M. A. Shah, and A. M. Khattak, “EEG-based emotion recognition: A systematic review of traditional and deep learning methods,” Premier Sci., vol. 2025, p. 1257, 2025, doi: 10.52700/pjs.v25.1257.
[9] D. Chicco and G. Jurman, “The advantages of the Matthews correlation coefficient (MCC) over F1 score and accuracy in binary classification evaluation,” BMC Genomics, vol. 21, p. 6, Jan. 2020, doi: 10.1186/s12864-019-6413-7.
[10] S. Agarwal, A. Sharma, R. Kumar, and P. Singh, “Machine learning evaluation of imbalanced health data: A comparative analysis of balanced accuracy, MCC, and F1 score,” in Proc. Int. Conf. Mach. Learn. Med. Imaging, vol. 15239, 2024, pp. 167–181, doi: 10.1007/978-3-031-72215-8_12.
[11] Y. Li, X. Chen, Z. Zhang, and L. Wang, “Real-time music emotion recognition based on multimodal fusion,” Alexandria Eng. J., vol. 113, pp. 448–458, Dec. 2024, doi: 10.1016/j.aej.2024.11.082.
[12] S. Li, R. Wang, and J. Zhang, “A novel signal channel attention network for multi-modal emotion recognition,” Front. Neurorobot., vol. 18, p. 1442080, Aug. 2024, doi: 10.3389/fnbot.2024.1442080.
[13] Z. L. Zahari, “An enhanced stress indices in signal processing based on advanced Matthew correlation coefficient (MCCA) and multimodal function using EEG signal,” Ph.D. dissertation, Faculty of Electr. Electron. Eng. Technol., Univ. Malaysia Pahang, Pekan, Malaysia, 2023.
[14] P. D. Welch, “The use of fast Fourier transform for the estimation of power spectra: A method based on time averaging over short, modified periodograms,” IEEE Trans. Audio Electroacoust., vol. 15, no. 2, pp. 70–73, Jun. 1967, doi: 10.1109/TAU.1967.1161901.
[15] J. A. Russell, “A circumplex model of affect,” J. Pers. Soc. Psychol., vol. 39, no. 6, pp. 1161–1178, Dec. 1980, doi: 10.1037/h0077714.
[16] R. J. Davidson, “Anterior cerebral asymmetry and the nature of emotion,” Brain Cogn., vol. 20, no. 1, pp. 125–151, Sep. 1992, doi: 10.1016/0278-2626(92)90065-T.
[17] P. Stoica and P. Babu, “Pearson–Matthews correlation coefficients for binary and multinary classification,” Signal Process., vol. 222, p. 109511, Sep. 2024, doi: 10.1016/j.sigpro.2024.109511.
[18] K. Osman, M. A. Mosleh, and A. H. Ali, “Applying the six-sigma methodology to determine the limits of quality control tests for a specific linear accelerator,” J. Appl. Clin. Med. Phys., vol. 25, no. 10, p. e14492, Oct. 2024, doi: 10.1002/acm2.14492.
[19] G. Varoquaux, “Cross-validation failure: Small sample sizes lead to large error bars,” NeuroImage, vol. 180, pp. 68–77, Oct. 2018, doi: 10.1016/j.neuroimage.2017.06.061.
[20] A. Vabalas, E. Gowen, E. Poliakoff, and A. J. Casson, “Machine learning algorithm validation with a limited sample size,” PLoS ONE, vol. 14, no. 11, p. e0224365, Nov. 2019, doi: 10.1371/journal.pone.0224365.
[21] R. J. Barry et al., “k-Fold cross-validation can significantly over-estimate true classification accuracy in common EEG-based passive BCI experimental designs,” Sensors, vol. 23, no. 13, p. 6077, Jul. 2023, doi: 10.3390/s23136077.
[22] S. Kumar, R. Sharma, and A. Verma, “Optimized EEG-based stress detection: A novel approach,” Biomed. Pharmacol. J., vol. 17, no. 4, pp. 2307–2318, Dec. 2024, doi: 10.13005/bpj/2989.
[23] D. Chicco, N. Tötsch, and G. Jurman, “The Matthews correlation coefficient (MCC) is more reliable than balanced accuracy, bookmaker informedness, and markedness in two-class confusion matrix evaluation,” BioData Min., vol. 14, p. 13, Feb. 2021, doi: 10.1186/s13040-021-00244-z.
[24] L. D. Brown, T. T. Cai, and A. DasGupta, “Interval estimation for a binomial proportion,” Stat. Sci., vol. 16, no. 2, pp. 101–133, May 2001, doi: 10.1214/ss/1009213286.
[25] X. Li, W. Zhang, and Y. Liu, “An automated extraction of spectral-temporal and spatial-temporal features of EEG for emotion detection,” Brain Inform., vol. 12, p. 265, Jan. 2025, doi: 10.1186/s40708-025-00265-y.
[26] M. A. Khan, A. R. Baig, and S. A. M. Al-Sarem, “A novel optimized hybrid deep learning framework for mental stress detection using electroencephalography,” Brain Sci., vol. 15, no. 8, p. 835, Aug. 2025, doi: 10.3390/brainsci15080835.
[27] H. Zhang, J. Zhu, and Q. Zhao, “A hybrid EEG-based stress state classification model using multi-domain transfer entropy and PCANet,” Brain Sci., vol. 14, no. 6, p. 595, Jun. 2024, doi: 10.3390/brainsci14060595.
[28] M. T. Bilal, S. A. Khan, and M. U. Rehman, “Optimal variational mode decomposition based automatic stress classification system using EEG signals,” Appl. Acoust., vol. 228, p. 110297, Dec. 2024, doi: 10.1016/j.apacoust.2024.110297.
[29] D. M. W. Powers, “Evaluation: From precision, recall and F-measure to ROC, informedness, markedness and correlation,” J. Mach. Learn. Technol., vol. 2, no. 1, pp. 37–63, 2011.
[30] W. Chen, X. Li, H. Liu, and Y. Zhao, “Emotion recognition based on multimodal physiological electrical signals,” Front. Neurosci., vol. 19, p. 1512799, Jan. 2025, doi: 10.3389/fnins.2025.1512799.
[31] Y. Wang, H. Li, X. Chen, and Z. Zhou, “Learner’s cognitive state recognition based on multimodal physiological signal fusion,” Appl. Intell., vol. 55, p. 5958, Dec. 2024, doi: 10.1007/s10489-024-05958-1.
[32] A. Ahmad, T. Khan, and S. Hussain, “Enhancing EEG-based emotion detection with hybrid models: Insights from DEAP dataset applications,” Sensors, vol. 25, no. 6, p. 1827, Mar. 2025, doi: 10.3390/s25061827.
[33] T. Li, Y. Chen, Z. Wang, and X. Sun, “Leveraging deep learning for robust EEG analysis in mental health monitoring,” Front. Neuroinform., vol. 18, p. 1494970, Jan. 2025, doi: 10.3389/fninf.2024.1494970.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 The Author(s)
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
