Predicting NIRF Ranking via Novel K-Star Ensembles

The National Institutional Ranking Framework (NIRF) was established by the Indian Ministry of Education (MoE), which ranks higher education institutions across the country. The institutions being ranked may benefit if their rankings or performance could be predicted given their parameters, thereby, improving their Institute’s attributes. However, less attention has been given to the problem of predicting rank from the parameters collected for the rankings. The main objective of this research is to forecast the ranking of the Institutions using a four-year dataset of NIRF collected for 16 Indian Institute of Technologies (IITs). The change in parameter data was collected across 16 parameters to predict the change in ranks using two consecutive years. A novel K-Star 2 -stage ensemble algorithm was developed. In the first stage, a number of lazy learning algorithms, K-Star, Instance-Based Learner (IBk) and Locally Weighted Learning (LWL), were trained using 10 -fold crossvalidation. The predictions of these algorithms from stage 1 were taken into stage 2, where the K-Star predicted the change in rank. The results revealed that the K-Star Ensemble had an RMSE of 0.009 compared to other algorithms including IBk (RMSE: 5.76), K-Star (RMSE: 5.85), and LWL (RMSE: 5.91). We highlight the main implications of our work for rank prediction in the real world.