Smart Education Analytics: Student Academic Performance Prediction Using Ensemble Of Deep Learning Models With Improved Pelican Optimization Algorithm

Abstract

Educational institutions, in this  era of Technological Advancements require advanced, precise innovations in order to deliverstudent-centric and quality education to learners. In spite of the vast amount of data accessible to educational institutions, there is still ashortage for accurate methods, which observes and evaluates academic performance of students to actively target helpful activities that would translate the attempts of students and educational institutions, into precise  analytical methodsby utilizing Student Intervention Strategies in a proactive manner. Educational Data Mining (EDM) is that domain of Artificial Intelligence (AI) that utilizes an integration of Data Mining (DM) and Machine Learning (ML) models for making forecasts on aspects particularly related to teachers and students, and, educational institutions, in general. In this manuscript, we offer a solution for Student Academic Performance Prediction Using an Ensemble of Deep Learning Models and Improved Pelican Optimization Algorithm (SAPPEDL-IPOA). The main objective of the SAPPEDL-IPOA technique is to provide a precise and strong predictive framework for enhancing the monitoring of students’ academicperformance assessments. At a primary stage, the SAPPEDL-IPOA model applies the min-max scaling approach for data pre-processing to ensure that all features are scaled within a specific range. Next, an ensemble of DL approaches, namely, Bi-directional Long Short-Term Memory (BiLSTM), Double Deep Q Networks (DDQN), and stacked AutoEncoder (SAE) model is utilized for classification of Students’ Academic Performance. Finally, the Improved PPelican Optimization Algorithm (IPOA) is used for hyper-parameter tuning, in orderto optimize the parameters of the Ensemble models to enhance classification performances. To validate the improved performance of SAPPEDL-IPOA model, an extensive range of simulations is implemented and the occurances and outcomes are studied under various aspects. The comparison of the investigations delineates the improvement of the novel SAPPEDL-IPOA system under various metrics.