Graduate Admission Chance Prediction

Inspiration

As all our teammates are current undergraduate juniors, and we all plan to apply for graduate school in this fall, we are somehow anxious about our performance and application result one year later. Through talking or chatting with friends around us in daily life, we discovered that almost everyone has the same concern with us.

What it does

By taking basic information of the user, such as GPA and GRE score, our model can give a prediction to the graduate admission probability. This model can be used to help applicants to check the chance of admit to some levels of colleges so that they can make a better choice on what colleges they should apply. As a result, applicants can get into their desired graduate program.

How we built it

Find graduate admission data online and prioritize the data from our own university;
Clean and wrangle the dataset and conduct an exploratory data analysis;
Build three machine learning models and tune their parameters into the best performance;
Work on a simple interactive demo for frontend.

Challenges we ran into

Time difference: Three out of four teammates are currently in China, and another one is in the US, so we have a big time difference and only have a few hours overlapped to work together. We collaborated on Google Drive and Colab to work on the paper and code;
First Experience: We are all new to Datathon and its culture. At the beginning of the project, we spent much time on finding appropriate and related datasets. If we prepared ahead, we could had more available time working on model building and testing.
Model Accuracy: At first we setup 8 classes of admitted chance. But after running all models, we realized that the accuracy was not high enough for general prediction even though the parameters were at the best stage. We did a few research online and utilized what we learned from ML classes, and it might be cause by too many classes. Then we reset our classification method and improved the model accuracy score.

Accomplishments that we're proud of

Finish this project with all desired goals in this short period of time;
High model accuracy and the model can be applied to widespread use;
The interactive window for providing personal information and chance of admit, which is what we didn't think about previously.

What we learned

Cooperation remotely with a great time difference;
How to find dataset efficiently and wrangle the data into what we want;
Enhance model performance from checking the data quality and model parameters;
The most important factor in graduate application learned from our model analysis, which gives us a hint for what we need to improve during rest of the months.

What's next for Graduate Admission Chance Prediction

Due to the time limit, we only find a dataset of 500 variables. We believe we can train our model in a better way if we have a larger and a more comprehensive dataset. On the other hand, we can try to apply other complicated and advanced models to our data, which may ends up a better accuracy score and perform a better prediction. Also, if our model is well constructed, we can try to publish our result by making a webpage. Graduate applicants can go to our website, provide their basic information, and get their chance of admit back as the output. In this way, our model can assist many applicants get a better result on their college admission.