Comparing 15 Machine Learning Classifiers for University Admissions: Why Random Forest Wins

Dataset and preprocessing

The underlying data is the Kaggle dataset "data-for-admission-in-the-university," comprising 400 examples across seven independent features — GRE Score, TOEFL Score, University Rating, Statement of Purpose, Letter of Recommendation, CGPA, and Research Experience — plus a continuous "Chance of Admission" score used to derive the binary target (Admission Status = 1 if Chance of Admission ≥ 0.80, else 0). The sample was expanded to 800 records via bootstrap resampling, then preprocessed using the dataprep Python library, which the author reports found the data free of outliers. The paper notes the target classes were imbalanced — the narrative text states 460 admitted vs. 340 not admitted, though the paper's own Figure 2 chart labels the same bars 430/370, an internal inconsistency in the source document rather than something introduced in this review. SMOTE was applied to correct for the imbalance before model training. The data was then split 80% training / 20% testing.

The 15-model comparison (training data)

Random Forest: training vs. test performance

Out-of-bag error for the Random Forest model was reported at 3.14%, and the confusion matrix on test data showed 83 true negatives, 12 false positives, 3 false negatives, and 72 true positives — consistent with high accuracy on both classes rather than the model simply favoring the majority class.

Evaluation metrics, defined

Accuracy = (TP + TN) / (TP + TN + FP + FN) — overall correctness, less informative on imbalanced data.

Precision = TP / (TP + FP) — of everyone predicted "admitted," what share actually was.

Recall = TP / (TP + FN) — of everyone actually admitted, what share the model caught.

F1-score — the harmonic mean of precision and recall, balancing both.

ROC-AUC — the area under the Receiver Operating Characteristic curve; measures how well the model separates the two classes across every possible decision threshold. Random Forest scored 0.99 — a score this close to the 1.0 maximum means the model almost never confuses an admitted applicant for a rejected one, or vice versa.

Feature importance

Random Forest's built-in feature-importance output ranked the seven inputs as follows (approximate share of total importance, read from the paper's Figure 3): CGPA ~30%, GRE Score ~22%, TOEFL Score ~18%, University Rating ~10%, Statement of Purpose ~9%, Letter of Recommendation ~7%, Research Experience ~3% (lowest of the seven).

Tools and libraries

The study was implemented in Python using Pandas, NumPy, Matplotlib, Seaborn, scikit-learn, dataprep (preprocessing), and PyCaret (automated model comparison). The final Random Forest model was serialized with joblib for deployment as a web application, API, or direct integration into an admissions system.

Limitations worth flagging

This is a single, moderately small public dataset (400 original records) from one unspecified admissions context, not a multi-institution or longitudinal study — the relative feature-importance ranking may not generalize across institutions, countries, or admissions cycles with different selection philosophies. The bootstrap-to-800 expansion increases sample size for model training but does not add genuinely new information beyond the original 400 records, and the two internal reporting inconsistencies noted above suggest the published version of the paper was not fully proofread against its own tables before submission.