Predictive modeling of student career pathways using machine learning techniques

Abstract

Career inappropriateness is common in Nigeria, leading to educational malcontent and labor-force ineffectiveness. Standard methods, even depending mainly on score-based assessments like JAMB, tend not to reflect multifaceted considerations affecting appropriate careers. Consequently, a machine learning model was designed to predict suitable career pathways with high accuracy by integrating academic performance (JAMB and WAEC/NECO scores) with personal attributes, including individual interests, skills, and extracurricular activities. Data of 3,231 students were utilized for investigation, and three models of prediction, i.e., Logistic Regression, Random Forest, and XGBoost, were trained from an 80:20 train-test split. Accuracy, precision, recall, and F1-score were utilized as measures of model performance, with SHAP analysis as an explainability tool. XGBoost had the best accuracy of 85.7%, where STEM careers were predicted with the best accuracy (92-94%) since academic criteria for STEM careers were definite. Personal attributes provided 28-39% of predictive power for non-STEM careers, where personal attributes emerge as crucial ingredients in combined counselor career crafting. These results suggest that combining academic and personal profiling has the potential to significantly outperform existing score-based methods in predicting suitable careers as routes and provide valuable insights for university admission committees, policymakers, and students as well. Extensions with longitudinal validations, integration with labor market data, and other populations and routes are proposed for enhanced applicability and national labor-force alignment.