A CRISP-DM and Predictive Analytics Framework for Enhanced Decision-Making in Research Information Management Systems

The age of digitization has led to a significant increase in the amount and variety of data, particularly within the research domain, where data previously stored in paper form has now been digitized and integrated into research management processes. The rapid growth of Big Data, driven by technologies like the Internet of Things, presents challenges for conventional data processing methods. However, data alone, stored in silos, lacks value. To unlock its potential, data must be analysed and processed to generate insights and predictions that enable evidence-based decision-making. Predictive Analytics (PA) is a powerful tool for this purpose. By leveraging PA and advanced statistical methods, predictive models for research management can be developed, helping to forecast research trends and outcomes, which in turn, provides decision-makers with a reliable, forward-looking basis for strategic decisions in research management. This paper explores the application of PA in Current Research Information Systems (CRIS) to enhance decision-making. A case study using metadata from 20,000 publications indexed in Scopus demonstrates how PA can identify emerging research topics and predict future trends. Machine learning algorithms such as Support Vector Machine (SVM), k-Nearest Neighbor (kNN), Random Forest, and Tree classifiers were employed, with metrics such as Area Under ROC (AUC), classification accuracy (CA), F1-score, precision, and recall evaluated. The results indicate that the kNN algorithm provided the highest performance with an AUC of 0.451 and a classification accuracy of 87.4%. These results show that predictive models can reveal significant patterns in research data, supporting data-driven decisionmaking for research management. Additionally, the study applied Latent Semantic Indexing (LSI) and clustering techniques to identify and categorize key topics within the data, showing a thematic focus on areas such as smart cities and urban intelligence before predictions, and CRIS applications after predictions. The findings illustrate how PA can optimize research management by identifying gaps in research and forecasting emerging topics, thereby aiding institutions in making more informed, evidencebased decisions