Comparative Evaluation of Machine Learning Classifiers for EEG-Based P300 Signal Classification Across Multiple Datasets

Abzal E. Kyzyrkanov, Zhenis Otarbay, Kumarova Assel, Nazira Tursynova, Algys Saltanat, Zauresh Yersultanova, Elmira Nurlanovna Tulegenova

2026

Full text Open access: Gold Metadata (CSL-JSON)
DOI 10.1016/j.procs.2026.08.401
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

Artificial intelligence-driven assistive technologies have increased interest in electroencephalography (EEG)-based brain–computer interfaces, where reliable Classification of the P300 event-related potential remains a central problem for communication and control systems. This study investigates whether a compact and transparent machine learning pipeline can provide strong P300 Classification performance across heterogeneous EEG datasets. Five datasets—NU, TenHealthyData, BNCI2015003, ALSdata, and EPFL—were analyzed using a unified workflow consisting of feature standardization, vectorization of multichannel EEG epochs, principal component analysis for dimensionality reduction, and supervised Classification with logistic regression, support vector classifier, random forest, k-nearest neighbors, AdaBoost, and multilayer perceptron models. The results show that the proposed comparative framework achieved strong performance across datasets, with the highest accuracy obtained on BNCI2015003, where k-nearest neighbors reached 96.5% and logistic regression reached 95.9%. TenHealthyData also produced high performance, with random forest reaching 92.2%, while random forest achieved the best reported result on the NU dataset with 85.2%. Across the full comparison, the findings indicate that classifier effectiveness is strongly dataset-dependent, but that carefully preprocessed EEG features can be classified accurately using both classical machine learning models and lightweight neural approaches. These results support the use of compact, interpretable, and computationally efficient pipelines for EEG-based P300 Classification in brain–computer interface applications relevant to assistive and pervasive computing.