Research basis

How the model was built

This application operationalizes the MSc dissertation "Predicting Campus Admission through Assessment of Soft Skills using Random Forest Algorithm."

Model configuration

Random Forest classifier · 500 decision trees · max depth 15 · minimum 4 samples to split a node · minimum 2 samples per leaf. Hyperparameters selected via grid search and cross-validation, as reported in the dissertation.

Current model performance

87.1%

Accuracy

0.879

Precision

0.879

Recall

0.879

F1 Score

0.97

ROC-AUC

Feature importance

Gpa

19.7%

Kcse Score

19.2%

Math Grade

18.6%

Science Grade

15.5%

Problem Solving Skill

8.2%

Leadership Skill

6.3%

English Grade

5.4%

Communication Skill

4.6%

School Type

0.9%

School Category

0.7%

Locality

0.5%

Gender

0.4%

The current training dataset is a synthetically generated dataset calibrated to reproduce the sample size (n=408), admission rate, group differences, and correlation structure reported in the dissertation's Chapter 4 results. Administrators can upload a real dataset via the admin dashboard to retrain the model on live data at any time.