Review and Analysis of Application of Improved Machine Learning Algorithms in Prediction of Students Academic Performance

Abstract

Access to higher education is essential for economic growth, social justice, and academic success.  However, dropout rates are a major issue for educational institutions worldwide.  Socioeconomic position is one of several factors that contribute to the wide range in dropout rates between countries.  Early identification of at-risk students is necessary to increase retention rates and carry out successful treatments.  This study predicts whether students will thrive academically or drop out using a variety of machine learning techniques.  From several years we evaluated demographic, socioeconomic, academic, social, and macroeconomic aspects of students in different majors.  Marital status, application mode, course, attendance type, prior qualifications, nationality, parental qualifications and occupations, special educational needs, gender, scholarship status, age at enrolment, debt status, tuition fee status, and curricular unit performance are among the 35 attributes that are included in the dataset.  In order to pre-process the data, pertinent classes and attributes were found, negative correlations were removed from features, and outliers were found and eliminated using the Interquartile Range (IQR) method.  We separated the dataset into a training set, which made up 67% of the total, and a testing set, which made up the remaining 33%, after normalizing it using Standard Scaler.  The hyperparameters were optimized via grid search.  Prediction models were developed using the following six classification algorithms: SVM, Decision Tree, Random Forest, Naive Bayes, K-Nearest Neighbours (KNN), and Logistic Regression.  It was discovered that the SVM model had the best F1-score, recall, accuracy, and precision.  Random Forest and Logistic Regression outperformed Naive Bayes, KNN, and Decision Tree.  The findings show that Random Forest, SVM, and Logistic Regression are effective models for predicting when students will leave school.  By providing schools with effective tools for early risk assessment and customized intervention strategies, this study emphasizes the value of machine learning in enhancing educational administration and enhancing student accomplishment.