High School Performance Based Engineering Intake Analysis and Prediction Using Logistic Regression and Recurrent Neural Network

Abstract

Abstract: A student's high school performance is crucial for engineering admission in Nepal. Machine

learning-based predictive models can provide valuable insights. This study aims to predict engineering entrance exam scores and admission probability based on high school academic records. In this study, we have used exam data from National Examination Board (NEB) and Institute of Engineering (IOE) containing grades, scores and results for over 11,000 students. Logistic Regression (LR) and Long-Short Term Memory (LSTM) models are implemented to predict pass/fail status and year-wise entrance score forecasting, respectively. In addition, the Prophet model analyzed trends in entrance score threshold averaging. The result shows that the logistic model achieved 97% accuracy in predicting pass/fail status and the LSTM network attained reasonable accuracy between 65-85% for score forecasting. The Prophet model accurately projected decreasing trends in threshold scores and admitted students' averages. Our model analyses provides actionable insights into student outcomes, complex patterns, and changing trends. Proactive interventions through upgraded curriculum, teacher training etc. could reverse declining enrolment.

Keywords: Education data mining, Intake prediction, Logistic regression, Long Short-Term Memory (LSTM), Student performance