Why GyanDhan is Launching the Admission Prediction Tool?

After completing his undergrad from one of the premier institutes in India, Jayant wanted to pursue higher studies in the US. He scored well in GRE, and he started his journey of looking for colleges who him best might accept his application. As the average time to complete an application for Top MS program in US is quite high, so it became important for him to shortlist the right colleges, as he wanted to put his best foot forward in all his applications.

"Someone must have surely solved this problem" he thought. After spending a whole day on one of the man's greatest inventions (saying internet is too mainstream) he was sure of 4 things:

1. Every admit predictor/ profile evaluator in the market classified colleges as Safe/Ambitious/Moderate. Since all the colleges falling in the same bucket (i.e. Safe/Ambitious/Moderate) are indistinguishable from each other, and several colleges fall in each of these buckets, hence it becomes really tough to make your list of 8-10 target colleges.
2. It seemed that all profile evaluators used average admit GRE scores for a given college to categorise it.
3. Most of the profile evaluators are not specialized for Indian students, and hence, give average results for all students globally.
4. No one solely relies on Profile evaluators – Everyone still uses forums, social media to get additional info before deciding their colleges shortlist.
   

Since Jayant was the Tech Lead at GyanDhan then, our team brainstormed as to how we could improve the status quo. And this is how our admit predictor was born.

We firmly believe that our Admit Predictor will significantly help students with their application process. Our belief stems from the following:

1. Our Admit Predictor is granular – we go beyond Safe/Ambitious/Moderate categorization by providing the admit probability for each college.
2. Every profile is computed using insights from thousands of historical data points.
3. Our percentile chart shows how you rank amongst the past successful applicants - so you can know whether you’re in the top 10% of the applicants to ASU MS in CS or in the top 30%!

Data Collection

Several students contributed to this initiative by sharing their info through surveys. In total, we reached out to 15,000 applicants in the GyanDhan community. Many students gave us access to Google spreadsheets, where past students had voluntarily entered their admit-related information.

Data Cleaning and EDA

After going through the dirty work of cleaning data, we normalized the continuous variables (GRE Score, CGPA, # of publications) and standardized the discrete variables (College Name, Branch, Work ex.)

Features Used

1. Acceptance ratio of top US colleges for Indian Students (computed from Analysis, ASEE)
2. Profile of student in Undergrad college :
   • Indian College Name
   • Branch
   • CGPA / Percentage
   • Rank of the student in the class (computed from Analysis)
3. Test scores:
   • GRE score (Quant, Verbal, AWA)
   • TOEFL/ IELTS
4. Target graduate course and college
   • US college name
   • Branch
   • Semester type ( Fall/ Spring)

Model Building

We tried different Machine learning models: logistic regression, SVM, random forest, and compared their results. After training our model and evaluating it on different test datasets, we concluded that a Hybrid Model (of Logistic regression and Random Forest) was outperforming the others. This is the model that runs in the background when you evaluate your profile on GyanDhan’s Admit Predictor.