SMORASO-DT : A hybrid machine learning classification model to classify individuals based on working memory load in mental arithmetic task

Shivabalan, K R; Deb, Brototo; Goel, Shivam; Arivan, R · 2020 · Crossref

DOI: 10.1101/2020.10.02.20205922

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the need for reliable, non-invasive methods to assess cognitive load and working memory performance, particularly in high-stakes environments like military operations or traffic control where decision-making errors can be critical. While subjective measures and physiological indicators like pupil diameter exist, they often lack objectivity or continuous monitoring capabilities. The authors propose using electroencephalography (EEG) signals, leveraging the brain’s nonlinear and chaotic dynamics, to differentiate between "good" and "bad" performers during mental arithmetic tasks. The primary goal is to develop a hybrid machine learning model capable of classifying individuals based on their working memory load, potentially serving as a biomarker for cognitive efficiency and early detection of cognitive impairments. The researchers utilized EEG data from the PhysioNet database, originally collected from 66 healthy participants performing mental arithmetic tasks. The methodology involved extracting both linear and nonlinear dynamic features from the EEG signals using EEGFrame software. Feature selection was conducted in two stages: first, a Random Forest algorithm identified the most significant variables based on minimal depth and importance metrics; second, Adaptive Lasso (ALASSO) with AICc validation further refined the feature set. The selected features were then input into machine learning classifiers, specifically a Fine Gaussian Support Vector Machine (SVM) and a Decision Tree, implemented in MATLAB with 7-fold cross-validation. The final hybrid model, termed SMORASO-DT (SMOte + Random Forest + Lasso-Decision Tree), was designed to handle class imbalance and optimize classification accuracy. The results demonstrated that the Decision Tree classifier outperformed the Fine Gaussian SVM, achieving an accuracy of 78% (reported as 79% in the conclusion) and an Area Under the Receiver Operating Characteristic Curve (AUROC) of 80%, compared to the SVM’s 73% accuracy and 73% AUROC. The analysis identified specific EEG features, particularly from the frontal lobe (e.g., EEG.Fp1.5, EEG.Fp1.10, EEG.Fp1.14), as statistically significant predictors. The study found that "good" performers exhibited lower entropy and fractal dimension in the theta and beta bands of the frontal lobe, indicating more regular and focused brain activity under high cognitive load. Conversely, higher entropy and fractal dimensions were observed in the occipital and temporal lobes for good performers. These findings suggest that increased mental workload correlates with decreased signal irregularity in frontal regions, reflecting enhanced cognitive focus. The significance of this work lies in the development of the SMORASO-DT model, which provides a robust framework for objectively assessing cognitive workload and performance. By integrating nonlinear dynamic features with advanced machine learning techniques, the study offers a more accurate classification of cognitive states than previous methods relying solely on linear features or principal component analysis. The authors conclude that this approach could be extrapolated to clinical settings for the early detection of dementia or minimal cognitive impairment, as well as for ergonomic applications to monitor and maintain productivity in high-workload professions. The study highlights the potential of EEG-based biomarkers to elucidate the complex dynamics of brain activity during cognitive tasks.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• neuro workload indices