Logistic Elitist Liquid Neural Network For Student Dropout Prediction

Abstract

Student dropout prediction involves detecting students who are at risk of exiting their studies before completion. This process utilizes the data on academic performance, engagement, demographic factors, and other relevant indicators to predict possible dropouts, allowing organizations to execute timely interventions to improve maintenance rates.Different machine learning and deep learning methods are developed for the early identification of students at risk of dropping out has gained a lot of interest recently. But, the accurate dropout prediction with minimal error is a major concern. A novel Logistic regressive Elitist Optimized Liquid Neural Network (LOGEO-LNNet) is developed for categorizing the student dropout with high accuracy and minimal error rate. The proposed LOGEO-LNNet consists of four major steps namely data acquisition, preprocessing, feature selection and classification. To begin with, the LOGEO-LNNet model performs data acquisition by collecting the studentdata samples from a dataset. Followed by, data preprocessing is employed which includes missing data imputation and outlier deletion. Then the combinatorial Target projection matching based feature selection process is employed for identifying the morerequired features. With these selected features, the system proceeds to the classification process by employing Liquid Neural Networks to distinguish the more than two categories by employing Polytomous logistic regression. Fine-tuning the layers of liquid neural network is a vital step using the adaptive elitist shuffled shepherd metaheuristic algorithm thereby minimizing errors and increasing the accuracy of the student dropout prediction. Experimental evaluation of LOGEO-LNNet model is carried out with different factors such as accuracy, precision, sensitivity, F1 score, specificity and student dropout predictiontime, confusion matrix with respect to a number of student data.The experimental result reveal that the proposed LOGEO-LNNet model consistently achieves superior student dropout prediction accuracy performance, exhibiting lower error rates and reduced prediction time compared to existing deep learning methods.