IVY League College Predictor

Linear Regression model to predict the chance of admission to IVY league college

See LIVE demo here

Project Organization

├── LICENSE            <- Open-source license if one is chosen
├── Makefile           <- Makefile with convenience commands like `make data` or `make train`
├── README.md          <- The top-level README for developers using this project.
├── data
│   ├── external       <- Data from third party sources.
│   ├── interim        <- Intermediate data that has been transformed.
│   ├── processed      <- The final, canonical data sets for modeling.
│   └── raw            <- The original, immutable data dump.
│
├── models             <- Trained and serialized models, model predictions, or model summaries
│
├── requirements.txt   <- The requirements file for reproducing the analysis environment, e.g.
│                         generated with `pip freeze > requirements.txt`
│
├── setup.cfg          <- Configuration file for flake8
|          
|__ EDA.ipynb          <- EDA notebook
|          
|__ src/train.py       <- training script
|
|__ streamlit_app.py   <- streamlit app

Intro to the Dataset and the Aim

An education institute has recently launched a dataset that contains the details of students who have applied for admission to IVY League College. The Jamboree team wants to know what factors are important for a student's success in getting into an IVY league college. They also want to see if we can make a predictive model to predict the chance of admission to IVY league college using the given features.

Dataset

This dataset contains the details of 500 students who have applied for admission to IVY League College along with their success rate.

Summary of sanitized data:

Column	Description
serial_no	Unique row ID
gre_score	Out of 340
toefl_score	Out of 120
university_rating	Out of 5
sop	Out of 5
lor	Out of 5
cgpa	Out of 10
research	Either 0 or 1
chance_of_admit	Ranging from 0 to 1

Additional feature engineered columns:

Column	Description
gre_sqr	Square of gre_score
cgpa_sqr	Square of cgpa
uni_rating_sqr	Square of university_rating
gre_uni_ratio	gre_score/university_rating
cgpa_uni_ratio	cgpa/university_rating
gre_cgpa_prod	gre_score*cgpa
gre_avg_uni_rating	Avg gre_score grouped by university_rating
cgpa_avg_uni_rating	Avg cgpa grouped by university_rating

Aim:

1. To analyze what factors are important for a student's success in getting into an IVY league college.
2. To make a predictive interpretable model to predict the chance of admission (chance_of_admit) to IVY league college using the given features.

Methods and Techniques used: EDA, feature engineering, modeling using sklearn pipelines, hyperparameter tuning

Measure of Performance and Minimum Threshold to reach the business objective: MSE of 1% or less with max VIF less than 5

Assumptions

1. This fairly small dataset (500 entries) is representative of the real-world population.
2. The data is stable and does not change over time. Thus model is assumed to not decay.

Results

• The best model is Linear Regression with
  • MSE: 0.3421%
  • Accuracy(R^2) 82.282%
  • VIF: 4.26
• The following features with weights are selected by the model which signifies the importance of those features Intercept: -0.25633387548676145

Features	Coefficients
cgpa	0.599586
research	0.205412
gre_score	0.122329
toefl_score	0.097854
university_rating	0.068220

• CGPA has the most weight for predicting the chance of admit to IVY league college followed by research and GRE score

Check EDA under /notebooks for more details or see the Kaggle Notebook here