Pradip Dharam

 

 

Objective

 DonorsChoose.org has started receiving almost 500,000 of project proposals each year for classroom projects in need of funding. Large number of volunteers are needed to manually approve or reject the project funding proposal.

 

As a result, there are three main challenges they need to solve:

• Scale current manual process to check up to 500,000 projects so that they can be posted as quickly and as efficiently as possible
• How to increase the consistency of project vetting across different volunteers to improve the experience for teachers
• How to focus volunteer time on the applications that need the most assistance

 

The goal is to predict whether or not a DonorsChoose.org project proposal submitted by a teacher will be approved, using the text of project descriptions as well as additional metadata about the project, teacher, and school.

 

Solution

Observations:

 +---------------+------------+----------+

|     Model     | Test Score | Test AUC |
+---------------+------------+----------+
| LSTM - Model1 |   127.6    |  99.05   |
| LSTM - Model2 |    66.0    |  92.32   |
| LSTM - Model3 |   77.62    |  92.36   |
+---------------+------------+----------+

• All three models are overfitting
• Model 1 looks good performer in comparison on other models
   

Github Jupyter Notebook:

https://github.com/pradipdharam/DonorsChoose.org--funding-approval-/blob/master/A25.%20Donor%20Choose%20-%20LSTM%20Models.ipynb

Citation: Assignment is provided by www.appliedaicourse.com