Explainable machine-learning integration of pre-task and preparatory EEG: revealing complementary predictive information for expert action reliability.

Code ↔ Paper

11 matches between paragraphs of the paper and lines of its authors' code, computed by the harvester (lexical-v1). Click a colored paragraph or line to see its counterpart.

The 11 matches

[1] § Materials and methods › Explainability and multiscale integration ↔ code/04_multiscale_integration.py, lines 213–306 · score 0.87 · ridge regression, Nested leave, preparation spectra, multiscale integration, outperform, R2
[2] § Materials and methods › Unsupervised behavioral decomposition ↔ code/01_behavioral_architecture.py, lines 116–145 · score 0.81 · depth centroid, lateral bias, fine precision, area compactness, zone reliability, Pearson
[3] § Materials and methods › Spectral features and nested cross-athlete learning ↔ code/04_multiscale_integration.py, lines 71–190 · score 0.78 · logistic regression, target zone, hit, transformed, weighting, class
[4] § Results › Pre-task alpha organization related to athlete-level reliability and compactness ↔ code/02_pretask_alpha.py, lines 1–18 · score 0.77 · PC1 PC2 placement, fronto central alpha, Pre task alpha, partial correlations, style, coherence
[5] § Results › Multiscale integration combined complementary predictive information ↔ code/04_multiscale_integration.py, lines 213–306 · score 0.72 · ridge regression, Nested leave, preparation spectra, regional, R2, alpha organization
[6] § Results › Dominant landing variance reflected individualized organization ↔ code/01_behavioral_architecture.py, lines 116–145 · score 0.70 · depth centroid, lateral bias, fine precision, zone reliability, PC3, components
[7] § Materials and methods › Spectral features and nested cross-athlete learning ↔ code/03_preparation_decoder.py, lines 34–47 · score 0.70 · logistic regression, feature selection, imputation, L2, weighting, Median
[8] § Results › Pre-task alpha organization related to athlete-level reliability and compactness ↔ code/02_pretask_alpha.py, lines 1–18 · score 0.68 · PC2 placement style, fronto central, pre task, coherence, IAPF, compactness
[9] § Results › Multiscale integration combined complementary predictive information ↔ code/04_multiscale_integration.py, lines 71–190 · score 0.65 · preparation coefficients, integrated model, combined model, pre task, outer, Multiscale
[10] § Materials and methods › Statistical analysis and software ↔ code/02_pretask_alpha.py, lines 119–138 · score 0.53 · controlled PC1, Partial correlations, Pearson, PC2, landing
[11] § Materials and methods › Experimental procedure ↔ code/01_behavioral_architecture.py, lines 1–16 · score 0.51 · unsupervised PCA, Split half, component, behavioral, landing, athletes

Paper

Loaded from Europe PMC by your browser, not stored by OSCR: doi.org · Europe PMC

The paper is loaded when this pane is shown.

The paper is loaded by your browser, with JavaScript: read it at doi.org or on Europe PMC.

The authors' code

Python · 357 lines · 13 KB · no license · 4 matches

#!/usr/bin/env python3
"""Integrate participant-level and trial-level predictive information.
Pre-task alpha organization supplies a participant-level baseline probability.
Preparatory EEG supplies within-participant trial evidence. The final analysis
combines those quantities without allowing the held-out participant to inform
model fitting, then tests whether averaged preparation features can reconstruct
the pre-task axis.
"""
import warnings
import numpy as np
import pandas as pd
from sklearn.feature_selection import SelectKBest, f_classif
from sklearn.impute import SimpleImputer
from sklearn.linear_model import LogisticRegression, Ridge
from sklearn.metrics import brier_score_loss, log_loss, roc_auc_score
from sklearn.model_selection import LeaveOneGroupOut, LeaveOneOut
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import OneHotEncoder, StandardScaler
from common import (
DATA,
RESULTS,
logit,
macro_auc,
sigmoid,
within_participant_z,
write_json,
)
REGIONS = {
"frontal": ("F7", "F3", "Fz", "F4", "F8"),
"central": ("C3", "Cz", "C4"),
"parietal": ("P7", "P3", "Pz", "P4", "P8"),
"occipital": ("O1", "O2"),
}
BANDS = ("theta", "alpha", "low_beta", "high_beta")
def one_hot_encoder():
"""Construct a dense encoder compatible with recent and older sklearn."""
try:
return OneHotEncoder(
drop="first", sparse_output=False, handle_unknown="ignore"
)
except TypeError:
return OneHotEncoder(
drop="first", sparse=False, handle_unknown="ignore"
)
def classifier(k, c):
"""Recreate the preparatory EEG classifier used in the decoder analysis."""
return Pipeline([
("impute", SimpleImputer(strategy="median")),
("scale", StandardScaler()),
("select", SelectKBest(f_classif, k=k)),
("model", LogisticRegression(
C=c,
penalty="l2",
class_weight="balanced",
solver="liblinear",
max_iter=4000,
)),
])
def integrate(preparation, predictions, pretask, selections, features):
"""Build pre-task, preparation, and combined held-out predictions.
The pre-task model estimates an athlete-level prior from alpha organization.
The preparation model estimates a common trial-level coefficient after
removing each participant's mean prediction. Their log-odds are added in
the combined model.
"""
base = preparation.merge(
predictions[["participant_id", "trial", "prediction"]],
on=["participant_id", "trial"],
how="inner",
).merge(
pretask[["participant_id", "alpha_organization"]],
on="participant_id",
how="left",
)
y = base["target_zone"].astype(int).to_numpy()
groups = base["participant_id"].to_numpy()
output = []
coefficients = []
# All components of the integration model are estimated without the outer
# test participant.
for train, test in LeaveOneGroupOut().split(base, y, groups):
participant = groups[test][0]
if base.iloc[test]["alpha_organization"].isna().all():
continue
inner = base.iloc[train].reset_index(drop=True)
inner_y = inner["target_zone"].astype(int).to_numpy()
inner_groups = inner["participant_id"].to_numpy()
inner_prediction = np.full(len(inner), np.nan)
# Refit the preparation decoder inside the outer training sample so the
# trial coefficient is based on cross-fitted, rather than fitted, scores.
for fit, validation in LeaveOneGroupOut().split(
inner, inner_y, inner_groups
):
choice = selections[
selections["held_out_participant"].eq(participant)
].iloc[0]
model = classifier(int(choice["k"]), float(choice["C"]))
with warnings.catch_warnings():
warnings.simplefilter("ignore")
model.fit(inner.iloc[fit][features], inner_y[fit])
inner_prediction[validation] = model.predict_proba(
inner.iloc[validation][features]
)[:, 1]
inner_state = within_participant_z(
logit(inner_prediction), inner_groups
)
encoder = one_hot_encoder()
participant_design = encoder.fit_transform(
inner_groups.reshape(-1, 1)
)
design = np.column_stack([inner_state, participant_design])
state_model = LogisticRegression(
C=1.0,
solver="liblinear",
class_weight="balanced",
max_iter=3000,
)
state_model.fit(design, inner_y)
state_beta = float(state_model.coef_[0, 0])
participant_training = (
inner.groupby("participant_id")
.agg(
hits=("target_zone", "sum"),
n=("target_zone", "size"),
alpha=("alpha_organization", "first"),
)
.dropna()
.reset_index()
)
participant_training["smoothed_rate"] = (
participant_training["hits"] + 0.5
) / (participant_training["n"] + 1)
scaler = StandardScaler()
x_trait = scaler.fit_transform(participant_training[["alpha"]])
trait_model = Ridge(alpha=1.0)
trait_model.fit(
x_trait, logit(participant_training["smoothed_rate"])
)
test_alpha = float(
base.iloc[test]["alpha_organization"].iloc[0]
)
trait_logit = float(trait_model.predict(
scaler.transform(pd.DataFrame({"alpha": [test_alpha]}))
)[0])
test_state = within_participant_z(
logit(base.iloc[test]["prediction"].to_numpy()), groups[test]
)
state_intercept = logit(
participant_training["hits"].sum()
/ participant_training["n"].sum()
)
model_predictions = {
"pre-task": np.repeat(sigmoid(trait_logit), len(test)),
"preparation": sigmoid(state_intercept + state_beta * test_state),
"combined": sigmoid(trait_logit + state_beta * test_state),
}
coefficients.append({
"held_out_participant": participant,
"preparation_coefficient": state_beta,
})
for model, values in model_predictions.items():
for row, value in zip(test, values):
output.append({
"model": model,
"participant_id": participant,
"trial": int(base.iloc[row]["trial"]),
"observed": int(y[row]),
"prediction": float(value),
})
return pd.DataFrame(output), pd.DataFrame(coefficients)
def summarize_integration(predictions):
"""Report discrimination and calibration for each information source."""
rows = []
for model, frame in predictions.groupby("model"):
y = frame["observed"].to_numpy()
predicted = frame["prediction"].to_numpy()
groups = frame["participant_id"].to_numpy()
_, scores = macro_auc(y, predicted, groups)
rows.append({
"model": model,
"pooled_auc": roc_auc_score(y, predicted),
"macro_auc": scores.mean(),
"brier_score": brier_score_loss(y, predicted),
"log_loss": log_loss(y, predicted),
"n_trials": len(frame),
"n_participants": frame["participant_id"].nunique(),
})
return pd.DataFrame(rows)
def cross_scale_reconstruction(preparation, pretask, behavior):
"""Test whether average preparation spectra reconstruct participant axes.
Regional band-power summaries are evaluated with nested leave-one-out ridge
regression. Negative held-out R-squared means that predicting the training
mean would outperform the attempted cross-scale reconstruction.
"""
feature_columns = [
column for column in preparation.columns
if column.startswith(("Prep1_", "Prep2_"))
]
participant = (
preparation.groupby("participant_id")[feature_columns]
.mean()
.reset_index()
.merge(
pretask[["participant_id", "alpha_organization"]],
on="participant_id",
)
.merge(
behavior[["participant_id", "zone_reliability"]],
on="participant_id",
)
.dropna()
.reset_index(drop=True)
)
engineered = pd.DataFrame({
"participant_id": participant["participant_id"]
})
all_features = []
alpha_features = []
for stage in ("Prep1", "Prep2"):
for band in BANDS:
for region, channels in REGIONS.items():
names = [
f"{stage}_{channel}_{band}_log_power"
for channel in channels
]
name = f"{stage}_{band}_{region}"
engineered[name] = participant[names].mean(axis=1)
all_features.append(name)
if band == "alpha":
alpha_features.append(name)
engineered["alpha_organization"] = participant["alpha_organization"]
engineered["zone_reliability"] = participant["zone_reliability"]
rows = []
alphas = (0.1, 1.0, 10.0, 100.0)
feature_sets = {
"all_bands": all_features,
"alpha_only": alpha_features,
}
for feature_set, columns in feature_sets.items():
for outcome in ("alpha_organization", "zone_reliability"):
y = engineered[outcome].to_numpy()
prediction = np.full(len(y), np.nan)
# The ridge penalty is selected inside each outer participant fold.
for train, test in LeaveOneOut().split(engineered):
best = None
for alpha in alphas:
inner_prediction = np.full(len(train), np.nan)
for fit_relative, validation_relative in LeaveOneOut().split(train):
fit = train[fit_relative]
validation = train[validation_relative]
model = Pipeline([
("scale", StandardScaler()),
("ridge", Ridge(alpha=alpha)),
])
model.fit(engineered.iloc[fit][columns], y[fit])
inner_prediction[validation_relative] = model.predict(
engineered.iloc[validation][columns]
)
error = np.mean((y[train] - inner_prediction) ** 2)
if best is None or error < best[0]:
best = (error, alpha)
model = Pipeline([
("scale", StandardScaler()),
("ridge", Ridge(alpha=best[1])),
])
model.fit(engineered.iloc[train][columns], y[train])
prediction[test] = model.predict(
engineered.iloc[test][columns]
)
r2 = 1 - np.sum((y - prediction) ** 2) / np.sum(
(y - y.mean()) ** 2
)
rows.append({
"feature_set": feature_set,
"outcome": outcome,
"n_participants": len(y),
"loso_r2": r2,
})
return pd.DataFrame(rows)
def main():
# 1. Load outputs from the previous scripts.
preparation = pd.read_csv(DATA / "preparation_eeg_features.csv")
predictions = pd.read_csv(
RESULTS / "preparation_out_of_sample_predictions.csv"
)
selections = pd.read_csv(RESULTS / "preparation_hyperparameters.csv")
pretask = pd.read_csv(RESULTS / "pretask_analysis_sample.csv")
behavior = pd.read_csv(DATA / "behavior_summary.csv")
features = [
column for column in preparation.columns
if column.startswith(("Prep1_", "Prep2_"))
]
# 2. Combine participant-level alpha organization with trial-level
# preparation evidence.
integrated, coefficients = integrate(
preparation, predictions, pretask, selections, features
)
# 3. Summarize discrimination and calibration for each information source.
summary = summarize_integration(integrated)
# 4. Test whether averaged preparation spectra reconstruct pre-task axes.
reconstruction = cross_scale_reconstruction(
preparation, pretask, behavior
)
# 5. Save results.
integrated.to_csv(
RESULTS / "multiscale_predictions.csv", index=False
)
coefficients.to_csv(
RESULTS / "multiscale_coefficients.csv", index=False
)
summary.to_csv(
RESULTS / "multiscale_summary.csv", index=False
)
reconstruction.to_csv(
RESULTS / "cross_scale_reconstruction.csv", index=False
)
print(summary.to_string(index=False))
print("\nCross-scale reconstruction")
print(reconstruction.to_string(index=False))
if __name__ == "__main__":
main()

04_multiscale_integration.py at commit 4d7e7b8, no license · at the source

Overview

Authors: Qiwei Zhao¹, Chenglin Zhou¹

School of Psychology, Shanghai University of Sport, Shanghai, China

Institutions: Shanghai University of Sport (China)

Journal: Frontiers in neuroscience, volume 20, article 1907378

Dates: received 12 June 2026; accepted 13 July 2026; published online 30 July 2026

Type: Research article · Language: English

License: CC BY

Identifiers: DOI 10.3389/fnins.2026.1907378 · PMID 42597548 · PMCID PMC13468886 · OpenAlex W7171860621

Open access: gold, a free copy (OpenAlex)

Status: code verified

Categories: EEG (modality), human (organism)

Methods: Spectral & time-frequency, Preprocessing, Connectivity, Statistics, Smoothing, state filtering, decompositions, Machine learning, Evoked potentials

Keywords: alpha oscillations, brain-behavior integration, EEG, explainable machine learning, motor expertise, preparatory activity, table tennis

Topic: Sport Psychology and Performance (Developmental and Educational Psychology, Psychology), according to OpenAlex

Funding: National Natural Science Foundation of China (32371129, 32500792); Key Lab of Exercise and Health Sciences of Ministry of Education (Shanghai University of Sport)

Citations: not cited yet (Europe PMC); 50 references in the paper

Abstract

Introduction: Expert motor behavior varies among athletes and across repetitive actions; however, conventional brain-behavior analyses can conflate these statistical levels. We characterized the behavioral architecture of expert table tennis serves and tested whether pre-task and action-preparatory EEG contribute information at different behavioral scales.

Methods: A total of 22 expert athletes completed 100 cue-directed serves with trial-resolved landing coordinates. Unsupervised principal component analysis characterized complete landing distributions. Eyes-closed pre-task EEG provided a hypothesis-informed alpha-organization index combining individual alpha peak frequency and fronto-central alpha coherence. EEG spectral power during action preparation was evaluated using nested leave-one-athlete-out learning, within-athlete permutation testing, grouped permutation importance, and multiscale probability integration.

Results: The two dominant landing components explained 57.2% of behavioral variance and primarily reflected reproducible individualized placement patterns, whereas a smaller component (7.4% explained variance) was more closely aligned with task-defined performance. Pre-task alpha organization was associated with higher target-zone reliability (r = 0.68, p = 0.002) and a smaller 95% landing area (r = −0.50, p = 0.034; n = 18). Preparatory EEG contained modest information that generalized trial outcomes to unseen athletes (macro AUC = 0.589; p = 0.0063), with predictive information concentrated in theta activity in both preparation stages, Prep2 high-beta activity, and parietal sensors. Athlete-average preparatory activities did not reconstruct pre-task alpha organization under leave-one-athlete-out validation.

Discussion: Pre-task and preparatory EEG carried complementary rather than interchangeable predictive information. Pre-task EEG was related to between-athlete performance propensity, whereas preparatory EEG offered weak but reliable within-athlete trial evidence. The value of the transparent machine-learning framework did not lie solely in achieving high prediction accuracy; rather, it was its capability to discover behavioral structure, test cross-athlete generalization, explain predictive information, and integrate neural evidence at a statistical level relevant to its operation.

Reproduced under the paper's license (CC BY), from the paper cited above.

Repository

Its files are read in the Code ↔ Paper reader above, with 11 matches between paragraphs and lines of code.

Qiwei-Zhao/pre-task-and-preparatory-EEG-of-expert-action-reliability

License: none: the authors keep all their rights

State: the link answers, verified on 27 September 2026

Evidence: files inventoried

Commit: 4d7e7b8196930937c653cdc4bd7c5c4e4abe9584, 11 July 2026

Languages: Python (5)

Size: 12 files, 5 scripts

Software Heritage: not archived

Found in: “Data availability statement”

Holds: README

Not found: license file, CITATION.cff, environment file, tests, continuous integration, documentation

Tools: NumPy (5 files), pandas (5 files), scikit-learn (4 files), SciPy (4 files)

Availability: 1 check, the latest on 27 September 2026: the link answers

27 September 2026: the link answers

6 files

code/01_behavioral_architecture.py, Python, 206 lines, 3 matches
code/02_pretask_alpha.py, Python, 166 lines, 3 matches
code/03_preparation_decoder.py, Python, 280 lines, 1 match
code/04_multiscale_integration.py, Python, 357 lines, 4 matches
code/common.py, Python, 60 lines
README.md, Text, 26 lines

At the source: github.com/Qiwei-Zhao/pre-task-and-preparatory-EEG-of-expert-action-reliability

The paper's code and data availability statement is in the Data section.

Tracing map

Proposed by the machine: these links were found in the paper and verified at the source, without human review. The map will receive a Zenodo DOI once one of the paper's authors has validated it with their ORCID.

What the map holds:

1 repository of the authors' code, each at its verified commit, with its license and how the link was found in the paper;
5 scripts, each with its path and the digest of its content;
11 matches between paragraphs of the paper and lines of the code (method lexical-v1);
neither the text of the paper nor the code itself.

Its JSON (tracing-map.json) is deposited on Zenodo with its DOI once the map is validated.

Data

No dataset and no data link were found in the paper.

Data availability statement

The de-identified data and analysis code required to reproduce the main findings are available on GitHub: https://github.com/Qiwei-Zhao/pre-task-and-preparatory-EEG-of-expert-action-reliability.

Reproduced under the paper's license (CC BY), from the paper cited above.

Versions

The history of this record: each version stored by the harvester or made by a correction of its authors or of the maintainers of its code, and what changed in its facts. The texts of the paper (its abstract, its availability statements) are not part of it; versions that changed only those are not listed.

Version 2, 28 September 2026

Funding: added National Natural Science Foundation of China: 32371129, 32500792; Shanghai University of Sport

Version 1, 27 September 2026: the first record

Recorded: type, language, journal, volume, pages, dates, 2 authors, 7 keywords, 48 references.

Cite

This paper

Zhao, Q., & Zhou, C. (2026). Explainable machine-learning integration of pre-task and preparatory EEG: revealing complementary predictive information for expert action reliability. Frontiers in neuroscience, 20, 1907378. https://doi.org/10.3389/fnins.2026.1907378

BibTeX

@article{zhao2026explainable, author = {Zhao, Qiwei and Zhou, Chenglin}, title = {{Explainable machine-learning integration of pre-task and preparatory EEG: revealing complementary predictive information for expert action reliability}}, journal = {Frontiers in neuroscience}, year = {2026}, month = jul, volume = {20}, pages = {1907378}, publisher = {Frontiers Media SA}, issn = {1662-4548}, doi = {10.3389/fnins.2026.1907378}, url = {https://doi.org/10.3389/fnins.2026.1907378}, pmid = {42597548}, pmcid = {PMC13468886} }

RIS

TY - JOUR AU - Zhao, Qiwei AU - Zhou, Chenglin TI - Explainable machine-learning integration of pre-task and preparatory EEG: revealing complementary predictive information for expert action reliability T2 - Frontiers in neuroscience J2 - Front Neurosci PY - 2026 DA - 2026/07/30 VL - 20 SP - 1907378 SN - 1662-4548 PB - Frontiers Media SA DO - 10.3389/fnins.2026.1907378 UR - https://doi.org/10.3389/fnins.2026.1907378 LA - en ER -

CSL-JSON

{ "id": "10.3389/fnins.2026.1907378", "type": "article-journal", "title": "Explainable machine-learning integration of pre-task and preparatory EEG: revealing complementary predictive information for expert action reliability", "container-title": "Frontiers in neuroscience", "author": [ { "family": "Zhao", "given": "Qiwei" }, { "family": "Zhou", "given": "Chenglin" } ], "container-title-short": "Front Neurosci", "volume": "20", "page": "1907378", "DOI": "10.3389/fnins.2026.1907378", "PMID": "42597548", "PMCID": "PMC13468886", "ISSN": "1662-4548", "publisher": "Frontiers Media SA", "URL": "https://doi.org/10.3389/fnins.2026.1907378", "language": "en", "issued": { "date-parts": [ [ 2026, 7, 30 ] ] } }

The tracing map gets a citation of its own once an author has validated it and it has a DOI.

Contribute

The authors of this paper can claim it, correct its record and validate its tracing map, and the maintainers of its code (its owner, or a public member of its organization) correct what it says of their repository; anyone signed in can ask for its removal. Every request goes to OSCR's own machine, which answers it; your account page follows them.

Sign in with ORCID to claim this paper as one of its authors, correct its record or validate its tracing map: when the paper's metadata lists your ORCID iD, you are recognized at once. Maintainers of its code: sign in with GitHub, then claim the repository on your account page.

Request its removal

To ask OSCR to remove this record, the copies of its authors' scripts or its tracing map, use the removal request page: signed in, you say who you are, what to remove and why, then review and confirm the request. Published rules decide every request (how).

Discussion, reproductions, activity

Discussion: questions and error reports about this paper and its code, from signed-in readers and its authors. It opens with sign-in.

Reproductions: reports from readers who ran the authors' code: what they reproduced, with which environment, commit and data. It opens with sign-in.

Activity: what happens around this paper: new versions of its record, its map's validation, discussions and reproductions. It opens with sign-in.