Abstract
The rapid advancement of scientific research has led to an exponential growth in the number of research papers being published across varous disciplines. This thesis addresses the challenge of identifying potential breakthrough papers amidst the rapid growth of scientific literature. Our definition of breakthrough papers focuses on their groundbreaking nature, as well as their characteristics of novelty, creativity, and transformative impact. To illustrate this, we examine Nobel prize-winning papers as the noteworkthy example of breakthrough papers. A data-driven approach is proposed utilizing machine learning and data mining techniques to analyze patterns in scholarly publications. A comprehensive dataset for Physics, Chemistry, and Biology from Web of Science (WOS) database is collected and standardized. The methodology utilizes features such as Impact, Potentiality, and Disruption while comparing them with traditional citation metric. Experiments and validation tests evaluate the framework's performance using Precision, Recall, F1-score, and AUC-ROC. The findings enable informed decision-making for researchers and stakeholders, while also fostering further exploration in computational scientometrics.