PredCheck: Detecting Predatory Behaviour in Scholarly World

An open automation system for predatory journal detection

“…In machine learning, it is generally assumed that the more pronounced the characteristics of classifier training, the better the results produced by a classification model. Unlike the text-based classification methods described in Bedmutha et al 26 and Adnan et al 46 , our proposed system uses diff scores (a measure of differences in specific word frequencies between journals) to identify feature word sets for classification prediction purposes. AJPC also provides objective data from three predatory journal lists: updated Beall’s, Stop Predatory Journals, among others.…”

“…In addition, maintaining a high F1 score does not sacrifice too much precision. Bedmutha et al 26 used research articles to training the model and got more than 0.71 F1 score in engineering area and 0.9 F1 score in biomedical area. Adnan et al 46 utilized Heuristic features to achieve an 0.98 F1 score, but this takes too much processing time compared to the NWF method.…”

“…As long as the overall input size is sufficiently large, its performance is considered suitably robust to handle parameter changes. In a study designed to detect instances of phishing, the RF classifier had a 98.8% accuracy rate 41 , and in a separate study aimed at detecting predatory biomedical journals, it produced an F1 score of 0.93 26 . The RF system has also been used with decision trees as a strategy for preventing the indexing of papers published in predatory journals since some individuals have become skilled at hijacking journal websites and collecting processing and publication fees from unwary authors 25 .…”

“…Data pattern extraction strategy for false indexing claim detection uses elements of random forest, RepTree, and J48 decision trees plus related algorithms 25 . To determine links between individual articles and predatory/legitimate publishers and journals, a data-driven training model, called PredCheck, was used with datasets from two India-based publishing groups: OMICS (OPG) and BioMedical Central (BMC) 26 . On average, their naive Bayes classifier-based model attained 95% accuracy and an F1 score of 0.89.…”