Can non-retracted published research articles be differentiated from research articles that are retracted due to error and misconduct?

Abstract

To retain and raise trust in science, it is essential to correct misinformation promptly, and even better to prevent the publication of incorrect information, to begin with. Taking a technical approach, this study attempts to address this critical issue of misinformation and trust in science by building models with the ability to classify retracted and non-retracted published scientific articles. These classifiers could be used by institutions to detect papers containing misinformation before they are published. Further, this study highlights the advantage of differentiating between scientific articles that have been retracted due to error and scientific articles that have been retracted due to misconduct. With this distinction, a Logistic Regression classifier was able to achieve an F1 weighted test score of 0.75 and an external validation score of 0.67.