Prediction of Academic Talent Capacity Based on Gradient Boosting Decision Tree

摘要

Talent introduction is an important force of academic development in universities. As the core of talent introduction, prediction of academic talent capacity is an essential and valuable research. However, it is hard to apply traditional statistical methods to extract knowledge from the mass and multi-dimensional talent information. Data mining approaches as up-to-date and efficient technologies are good at analyzing information, extracting patterns or rules from a big dataset and then making a prediction based on the relationship among extracted information. In this study, a series of data mining approaches are employed to evaluate the academic capacity of talent and to analyze the correlation between features. The Principal Component Analysis and Random Forest are used to feature extraction for improving the accuracy of prediction. A classical classification model, Gradient Boosting Decision Tree, is used as the primary analytic model to prediction. In order to validate the effectiveness of the model, other five classification models are used to conduct a comparative experiment based on prediction accuracy values and the F-measure metric. Further, to investigate the contribution of some important features, we make a marginal utility analysis of important features which have a high correlation with academic talent capacity. The experiment results reveals the important features for academic capacity and the positive factors for the academic production of talents.