Machine Learning Identification of EEG Predictors of Load-Specific Strength Gains Following Alpha Neurofeedback in Elite Judokas

This dataset accompanies the study "Machine Learning Identification of EEG Predictors of Load-Specific Strength Gains Following Alpha Neurofeedback in Elite Judokas", which investigated whether baseline alpha-band EEG metrics can predict individual responsiveness to neurofeedback in the context of strength adaptation. Conducted on 24 elite male judokas, the study implemented a 15-session alpha neurofeedback intervention targeting frontal alpha asymmetry (FAI) with EEG recorded from F3 and F4 electrode sites. Strength outcomes were assessed pre- and post-intervention across five loading intensities (35–100% 1RM) using a standardized back squat protocol. The research was structured in two analytical stages. First, multivariate linear regression models identified load-specific EEG predictors (FAI, F3, F4) of strength gain. Principal Component Analysis (PCA) supported the dimensionality reduction of EEG data prior to modeling. Second, supervised machine learning classifiers Random Forest (RF) and Multi-Layer Perceptron (MLP), were trained to categorize athletes as responders or non-responders based on baseline EEG features and training status. Both classifiers demonstrated above-chance accuracy, with F4_PRE and FAI_PRE emerging as the most influential features. The dataset, titled "EEG Alpha Neurofeedback and Strength Adaptation – Judokas (2024)", contains preprocessed and standardized EEG metrics (F3_PRE, F4_PRE, FAI_PRE), delta strength outcomes (ΔReps per %RM), classification labels, and all R scripts used for regression, PCA, and machine learning analysis. EEG preprocessing included ICA artifact removal, z-score normalization, and amplitude thresholding (±75 µV). The entire analysis pipeline follows open science and FAIR principles. This is the first study to integrate PCA-regression and machine learning to derive EEG-based predictive markers of neuromuscular adaptation in elite strength athletes. The results support the application of individualized neurofeedback strategies guided by neurophysiological traits and establish a foundation for precision brain–performance interfaces in sport science.本数据集配套于题为《优秀男子柔道运动员α波段神经反馈后负荷特异性力量增益的脑电图预测因子机器学习识别》的研究，该研究旨在探究基线α波段脑电图（Electroencephalogram, EEG）指标能否预测力量适应背景下个体对神经反馈的响应程度。本研究招募24名优秀男子柔道运动员，实施了为期15次的α波段神经反馈干预，干预靶点为额叶α波不对称性（Frontal Alpha Asymmetry, FAI），通过F3与F4电极点位采集脑电图信号。研究采用标准化背蹲起方案，在干预前后针对5种负荷强度（35%~100% 1RM）评估力量表现。 本研究采用两阶段分析框架：第一阶段，通过多元线性回归模型识别力量增益的负荷特异性脑电图预测因子（FAI、F3、F4）；建模前采用主成分分析（Principal Component Analysis, PCA）对脑电图数据进行降维处理。第二阶段，基于基线脑电图特征与训练状态，通过监督式机器学习分类器——随机森林（Random Forest, RF）与多层感知机（Multi-Layer Perceptron, MLP）——将运动员划分为响应者与非响应者。两类分类器均表现出高于随机猜测的分类精度，其中F4_PRE与FAI_PRE成为最具影响力的特征。 本数据集题为《脑电图α神经反馈与力量适应——柔道运动员（2024）》，包含经过预处理与标准化的脑电图指标（F3_PRE、F4_PRE、FAI_PRE）、力量变化量指标（每%RM对应的重复次数变化量ΔReps）、分类标签，以及用于回归、主成分分析与机器学习分析的全部R脚本。脑电图预处理流程包括独立成分分析（Independent Component Analysis, ICA）伪迹去除、Z分数标准化以及幅值阈值限定（±75 µV）。整套分析流程遵循开放科学与FAIR（可发现、可获取、可互操作、可复用）原则。 本研究首次将主成分分析-回归与机器学习相结合，从优秀力量项目运动员的脑电图信号中提取神经肌肉适应的预测标志物。研究结果支持基于神经生理特征的个性化神经反馈策略应用，并为运动科学领域的精准脑-表现交互界面奠定了基础。