Predicting the continuity of international scientific collaboration in the BRICS: a machine learning approach

Abstract

This study aims to predict respondent-reported continuity of international scientific collaboration within the BRICS through a supervised binary classification approach. It examines whether respondents reported that collaborations continued in the form of joint projects and co-authored publications, using survey-based variables related to collaboration climate, partner attributes, perceived benefits, barriers and impacts. The analysis is based on survey data comprising 313 valid responses for collaborative projects and 304 valid responses for publications involving researchers engaged in cross-border cooperation within the BRICS. The machine-learning input variables were derived from the survey responses; publicly accessible bibliometric and institutional sources were used only to identify eligible respondents. Seventeen variables were considered, including 15 predictors and two binary outcomes (Projects and Publications). Eight supervised machine learning classifiers were applied: Logistic Regression, support vector machine, k-nearest neighbors, decision tree, random forest, gradient boosting, histogram-based gradient boosting and naïve Bayes. Model performance was evaluated using stratified cross-validation and ShuffleSplit resampling, considering accuracy, F1-score, balanced accuracy and AUC metrics. The results indicate that trust culture is the most consistent predictor of respondent-reported continuity across both outcomes, while resource availability and agenda alignment are particularly important in predicting respondent-reported project continuity. This study offers an interpretable, respondent-level classification of reported collaboration continuity in BRICS-internal scientific partnerships. Its originality lies in treating the reported continuity of collaboration as a knowledge management -relevant outcome, distinguishing project-based from publication-based continuity and showing which established collaboration conditions remain informative across these two forms of sustained knowledge relationships.