Interpretable Machine Learning for Research Evaluation: Transparent Bibliometric Rules and the Limits of Association with Institutional Recognition in Neuroscience

The rapid growth of the scientific literature has intensified interest in how bibliometric information relates to scientific recognition, particularly in interdisciplinary fields such as neuroscience. Existing studies often rely on isolated indicators or black-box models. Here, the outcome is defined narrowly as institutionally recognized influence, operationalized as receipt of a prestigious neuroscience award from one of five societies (SFN, IBRO, FENS, CNS, ANS), rather than scientific influence in general. We compute 64 author assessment parameters in four categories for a balanced dataset of 1060 researchers (530 awardees, 530 non-awardees), using publications dated before the award year. A multilayer perceptron with leave-one-out feature ablation (MLP-FIS) ranks the parameters, and the top five per category feed Gini-based decision trees from which human-readable rules are extracted. On a random 20% test split the highlighted rules capture 33–49% of awardees at 83–95% specificity. Under repeated cross-validation with in-fold feature ranking, trees reach 66–70% accuracy against 68–73% for random forests, and the MLP-FIS top-five sets were unstable across folds. On a temporal hold-out of 106 researchers, rule accuracy was 52–67%. The benchmark results show limited incremental predictive advantage over strong conventional alternatives; accordingly, the contribution is interpretive rather than performance-based. The rules are descriptive benchmarks of bibliometric patterns associated with institutional recognition, not validated tools for judging scientific influence or individuals, and the analysis is subject to citation-timing leakage, award-selection bias and data-quality problems.

Authors

Institutions

Publication Details

Journal
Computation
Published
2026-11-19
DOI
https://doi.org/10.3390/computation14110300
Primary Topic
scientometrics and bibliometrics research
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Interpretable Machine Learning for Research Evaluation: Transparent Bibliometric Rules and the Limits of Association with Institutional Recognition in Neuroscience

Mirtha Silvana Garat de Marín, Isabel de la Torre Díez, Eduardo Silva Alvarado, Ahmad Sami Al‐Shamayleh et al.
Computation
scientometrics and bibliometrics research
article

Interpretable Machine Learning for Research Evaluation: Transparent Bibliometric Rules and the Limits of Association with Institutional Recognition in Neuroscience

Mirtha Silvana Garat de Marín, Isabel de la Torre Díez, Eduardo Silva Alvarado, Ahmad Sami Al‐Shamayleh, Hammad Abbasi, Ghulam Mustafa, Muhammad Tanvir Afzal
article en

Abstract

The rapid growth of the scientific literature has intensified interest in how bibliometric information relates to scientific recognition, particularly in interdisciplinary fields such as neuroscience. Existing studies often rely on isolated indicators or black-box models. Here, the outcome is defined narrowly as institutionally recognized influence, operationalized as receipt of a prestigious neuroscience award from one of five societies (SFN, IBRO, FENS, CNS, ANS), rather than scientific influence in general. We compute 64 author assessment parameters in four categories for a balanced dataset of 1060 researchers (530 awardees, 530 non-awardees), using publications dated before the award year. A multilayer perceptron with leave-one-out feature ablation (MLP-FIS) ranks the parameters, and the top five per category feed Gini-based decision trees from which human-readable rules are extracted. On a random 20% test split the highlighted rules capture 33–49% of awardees at 83–95% specificity. Under repeated cross-validation with in-fold feature ranking, trees reach 66–70% accuracy against 68–73% for random forests, and the MLP-FIS top-five sets were unstable across folds. On a temporal hold-out of 106 researchers, rule accuracy was 52–67%. The benchmark results show limited incremental predictive advantage over strong conventional alternatives; accordingly, the contribution is interpretive rather than performance-based. The rules are descriptive benchmarks of bibliometric patterns associated with institutional recognition, not validated tools for judging scientific influence or individuals, and the analysis is subject to citation-timing leakage, award-selection bias and data-quality problems.

ComputationVol. 14(11)
Al-Ahliyya Amman University (JO), Universidad de Valladolid (ES), Shifa Tameer-e-Millat University (PK), University of Applied Sciences Potsdam (DE), Fundación Universitaria Colombo Internacional (CO), Universidad Europea del Atlántico (ES), Universidade Internacional do Cuanza (AO), Universidad Internacional Iberoamericana (PR), Universidad Internacional Iberoamericana (MX), Universidad de la Romana (DO)
scientometrics and bibliometrics research
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon