Automated Student Merit Prediction using Machine Learning

Imparting higher education to today’s generation is gaining in importance. With the National Education Policy 2020 put on the table, there has been a massive transformation in the education system. The Government is already investing a considerable sum in budgetary allocation in quality education. However, the Government’s effort in this regard could hardly bear fruit by itself. According to this survey of private engineering institutions under MAKAUT, the performance of the present generation is deteriorating rapidly, which will be a detrimental effect on the long-term viability of these institutions, as placement would be directly affected by this level of performance. The present study addresses the problem from three angles. First, Pearson’s correlation coefficient has been used to measure the correlation between the starting and closing ranks across private engineering college departments under the MAKAUT umbrella. Second, a multiple regression has been run to study the impact of the passage of prior periods on the current merit level. Third, a random forest model was fitted to predict the ranks for the 31 college departments in the test datasets based on 600 data points in the training dataset for 2020. Finally, a Chi-square test has been run with the predicted and actual values first and then on the smaller subset of the test data. For this purpose, data on WBJEE ranks of students admitted to private engineering colleges under MAKAUT from 2015 to 2020 have been used. Each of the opening and closing ranks of candidates has been subtracted from the rank 1 lakh to generate the desired set of data values. Analysis based on the Pearson correlation coefficient shows that candidates’ opening and closing ranks are correlated, college and department-wise. It perhaps gives evidence that admissions move in homogeneous cohorts of positions. Further, the estimates of the parameters from the multiple regression model suggest that the parameter estimates’ coefficients are negative or minimal, indicating negative to no effect of previous years’ figures on 2019 values. The predictions by the random forest model are reasonably close to the actual.