Volume 18, No. 6, 2021
ML Use For Forecasting The NIRF Ranking Of Engineering Colleges In IndiaAnd PCA To Find The Correct Weightage For The Best Result
Dr. Bishnu Prasad Mishra , Dr. Ratikanta Dash , Dr. Banashri Rath
Abstract
Evaluation provides factual position of any person or organization in the long queue. Annual evaluation for all participating institutions is carried out with respect to particular scalefor global as well as local ranking. National Institutional Ranking Framework (NIRF) is adopted in India to rank institute of eminence in various sectors such as University, Engineering, Management, Pharmacy, College, Medical, Law, Architecture and Dental. TRL, RP, GO, OI and Peer Perception are the criteria used to evaluate ranking with 30%, 30%, 20%, 10% and 10% weightage respectively. All above parameters are further sub-divided into 17 parameters with different sub-weightage. Some parameters have strong points for rejecting the application. In this paper, Engineering ranking factual data are considered for developing regression model using ML under individual heading. Multiple regression model has also been developed to cross-check the accuracy of the model. It is felt that the weightage given to each major 5 components needs verification. Machine Learning model using Python software has been developed to train computer to forecast the rank of any participating institution. Training has been imparted to this ML software with 80% random data from NIRF rank list. Subsequently, testing has been done with 20% test data from same NIRF rank list. A few numbers of test datahas been fed to the system and accordingly, accurate prediction has been made. Many findingsfrom this ML plots need further detailed interpretation and discussion for refinement of the weightage. Further, the overall combined evaluation has been studied using PCA. The synergy components under different Principal Components have been computed along with their contribution towards the overall evaluation and final ranking. The result encourages the scope for changing the weightage of five main components and modify the weightage for increasing accuracy of the evaluation process. The weightage can be altered and evaluation process can be made more accurate. ML can be used for forecasting the ranking of any interested institution with the correct input. On-line evaluation of NIRF ranking will be feasible once the parameters are finalized. ML plots show that the scattered plot is not evenly spread. This implies that the funds availability and spending capacity of the institutions are not at par with each other. Many institutions scoring zero in peer perception have obtained ranks within top 200. This needs to be re-examined. The PCA highlights that peer perception weightage criteria is correct and the teaching learning resources weightage is to be reduced for a better judgement. Graduation outcome, Outreach and Industries criteria are not provided with proper weightage. The lower correlation at PC3 and PC2 reveals above facts.
Pages: 5222-5233
Keywords: NIRF, PCA, ML, Weightage, TRL, RP, GO, OI, Peer Perception, Python, Training, Testing