Educational Machine Learning For College Recommendation and Career Alignment Among Incoming University Students

Abstract

This work created and tested a machine learning approach to help incoming university students find colleges that fit both education plans and career goals. It used school records for 7,000 students at Northwestern University, covering the years 2020 through 2024. The analysis looked at demographic data, school results, admission details, and career-related items. These factors were treated as possible signals for predicting college placement. Six supervised learning models were trained and compared: Multinomial Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, XGBoost, and LightGBM.  To judge each model, the study used accuracy, precision, recall, and F1-score. It also used five-fold cross-validation to reduce the risk of one split giving a misleading result. Among the models, Random Forest reached the best holdout accuracy at 80.79%. Its F1-score was 80.77%, which was also the top value for that metric. Still, Multinomial Logistic Regression looked more stable across folds. It posted the top cross-validation F1-score at 79.13%. It also did not show clear signs of overfitting. For that reason, it was described as the most sound choice for use in the institution. The study also ran SHapley Additive exPlanations (SHAP) to interpret the drivers of the predictions. SHAP suggested that Senior High School strand carried the most useful signal. Next came admission scores for English, then Science, then Mathematics. Overall, the results claim that educational machine learning can turn existing student data into clearer and evidence-based suggestions. The framework is meant to act as decision support. It should work alongside professional academic advising, respect student choices, and encourage careful use of artificial intelligence in higher education. It is presented as responsible, understandable, and centered on students.