Neural_Network_Charity_Analysis

Overview of Analysis

I received a CSV containing more than 34,000 organizations that have received funding from Alphabet Soup. I need to use the features in the dataset to create a binary classifier that is capable of predicting whether apllicants will be successful or not if they are funded by Alphabet Soup, using my knowledge of machine learning and neural networks. I will preprocess the data for a neural network model, train/evaluate the model, and try to optimize the model.

Purpose

The purpose of this analysis is to design a binary classifier that is capable of predicting whether or not applicants will be successful with funding from Alphabet Soup.

Results

Data Preprocessing

• The variable I am targeting is the IS_SUCCESSFUL column because it contains binary information that says whether or not funding was successful or not.

• The columns I am using as features for the model are APPLICATION_TYPE, AFFILIATION, CLASSIFICATION, USE_CASE, ORGANIZATION, STATUS, INCOME_AMT, SPECIAL_CONSIDERATIONS, and ASK_AMT.

• I removed the identification columns (EIN and NAME) because they are neither targets or features.

Compiling, Training, and Evaluating the Model

• For my neural network model I used two hidden layers, one with 24 nodes and one with 12 nodes. I used the ReLU activation function for the two hidden layers, and Sigmoid for the output layers because the output is a binary classification.

  • 

• I was not able to achieve the target model performance of 75%. This model does not accurately predict the outcome of charity donations.

• To improve the performance of the model, I tried three changes.

  • First, I added more nodes to the two hidden layers. Instead of 24 and 12 I used 80 and 40. The model was still under 75%.
    • 
  • Next, I tried to add more hidden layers, and I increased from two to three, with 80, 40 and 20 nodes respectively. This change didn't increase the model's performance to be greater than 75%.
    • 
  • The last thing I tried was using a different activation function. I used ReLU, Tanh, and Sigmoid, instead of only Relu for the hidden layers. This, once again, did not increase the model's performance to be over 75%.
    • 

Summary

I tweaked my model three times, and was not successful in reaching the target model performance of 75%. A better model I could use to potentially reach a target performance of 75% is Random Forest Classifier, since we are working with a binary classification problem. The Random Forest Classifier is a supervised machine learning model, which might have better results than my deep-learning neural network.