Academic Success Prediction Model
Academic-success prediction (Graduate/Dropout/Enrolled) from 24 student features, comparing 6 ML models (RF/XGBoost best, ~63%) — CSE422 AI project.
Who might need support?
Scroll to watch it unfold. Each beat is one documented step of the project.
- 01
24 features
Socio-economic, demographic and academic data for 4,424 students.
- 02
Prepare
Clean, encode and scale the features.
- 03
Compare models
Six classifiers, including a Keras neural network and Random Forest, are trained head to head.
- 04
Predict
Each student is classed as Graduate, Dropout or Enrolled.
- 05
Outcome
Graduate, Dropout or Enrolled, predicted early enough to help.
An illustration of how the system works, drawn from the write-up below. It is not recorded output.
Overview
A machine-learning study that predicts student academic outcomes — Graduate, Dropout, or Enrolled — from 24 socio-economic, demographic and academic features, so institutions can spot at-risk students early and target support. Built for CSE422 (Artificial Intelligence) on a 4,424-student dataset.
Approach
After cleaning, encoding and scaling the features (with EDA on correlations, outliers and class balance), six classifiers are trained and compared head-to-head — a Keras neural network, Random Forest, XGBoost, K-Nearest Neighbours, Decision Tree and Logistic Regression — each evaluated with accuracy, precision/recall/F1, confusion matrices and ROC/AUC.
Results
The ensemble methods came out on top — Random Forest and XGBoost at ~63% accuracy, with the neural network close behind (~62%) — on a genuinely hard three-class problem with class imbalance. The head-to-head comparison highlights where feature engineering and ensembling help most, and where minority-class prediction still struggles.
By the numbers
Tech stack & key skills
Core tools, methods and skills demonstrated in this project: