Charity Funding Predictor

Nonprofit foundation Alphabet Soup is looking for a tool to predict whether applicants they fund will be successful.

The neural network model being built is based on a CSV containing data on organizations that have received funding. Included columns:

• EIN and NAME - Identification columns
• APPLICATION_TYPE - Alphabet Soup application type
• AFFILIATION - Affiliated sector of industry
• CLASSIFICATION - Government organization classification
• USE_CASE - Use case for funding
• ORGANIZATION - Organization type
• STATUS - Active status
• INCOME_AMT - Income classification
• SPECIAL_CONSIDERATIONS - Special considerations for application
• ASK_AMT - Funding amount requested
• IS_SUCCESSFUL - Was the money used effectively

Step 1: Data Preprocessing

Steps 1 and 2 take place in the fund_predictor notebook. The specified steps were followed:

1. Read in the CSV and remove EIN and NAME columns
2. Determine number of unique values in each column
3. Using cutoff values, bin "rare" categorical values into category "Other" for columns APPLICATION_TYPE and CLASSIFICATION
4. Encode categorical variables with Pandas.get_dummies()

Step 2: Compile, Train, and Evaluate the Model

The stated steps were followed:

1. Create a neural network using tensorflow.keras
2. Add two hidden layers with the "ReLU" activation function
3. Add an output layer with the "sigmoid" activation function
4. Compile and train the model, saving the model's weights every 5 epochs
5. Evaluate the model's loss and accuracy with test data
6. Save the model to an HDF5 file

Step 3: Optimization

Optimization takes place in a separate notebook called AlphabetSoupCharity_Optimization.

The goal of optimization was to reach a target predictive accuracy higher than 75%. This goal was achieved by reintroducing the NAME column and reducing the number of nodes active in each hidden layer.

Step 4: Analysis

Through the optimization process, the accuracy of the model increased from around 72% to just under 79%.

The full analysis report is available to read at the link.