A group research project for the course ECE 143: Programming for Data Analysis as of Winter 2023. Our proposed topic was to perform data analysis on factors contributing to admission rates for graduate school. However, this idea expanded to developing an AI prediction system for graduate school admission, as it worked in conjunction to the data analysis. Our motivation behind this project was to highlight the stressful process of college applications. A prediction system geared for college admissions could help alleviate this pressure by saving time and enabling strategic planning.

The project was boiled down to cleaning the datasets, performing exploratory data analysis, employing and evaluating various base models, implementing feature selection and engineering, and optimizing and evaluating the final model. From our analysis, we found that features, such as GRE scores and cumulative GPA, have significant importance to admission; meanwhile, other factors, such as Research and Letters of Recommendation, are miniscule. The most appropriate model for this application was an ANN, yielding a ~93% AUC score.

This project easily needs further refinement (with more unseen data) before any aspirations in deployment. Although our model supports +1000 graduate schools, the employed dataset was heavily imbalanced, susceptible to overfitting. Additionally, the entire decision-making among college applicants is just complicated 🙃... There are many confounding factors/biases that dictate the decision-making of college admission readers. For now, this project is a great proof of concept and has potential to serve unique use-cases from identifying areas of improvement in student applications to automating the decision-making for college admission readers.