Alphabet Soup Charity Application Deep Learning Model

Overview of the analysis:

The analysis used real world dataset which contains more than 35k organizations` data, funded by Alphabet Soup in the past. With the help of this data, the aim of the analysis is to create a deep learning neural network model to predict the success/failure of the future funds.

Results:

Data Preprocessing

• The column named "IS_SUCCESSFUL" is target of this model (dependent variable)
• During optimization attempts, different feature sets were used to understand their effects on the loss/accuracy.
• For all optimization attempts "EIN" and "NAME" variables were removed.

Compiling, Training and Evaluating the Model

• How many neurons, layers, and activation functions did you select for your neural network model, and why?

Main neural network model has contained two hidden layers with 80 and 30 neurons respectively. The hidden layes used "relu"activation function while the output layer used 'sigmoid'activation function. In this analysis, the epochs number was 100.

• Were you able to achieve the target model performance?

The original neural network model could not achieve the target model performance which is 75%. The accuracy of this model was approximately 73% and the loss was 56%.

• What steps did you take in your attempts to increase model performance?

  Three different attempts were made to optimize the model:

  • Attempt #1 In this attempt, couple of different alterations were made to model. Two more variables were dropped ('STATUS' and 'SPECIAL_CONSIDERATION'), binning structures of 'APPLICATION_TYPE' and 'CLASSIFICATION' variables have been changed, a new hidden layer was added with 60 neurons. After all these changes, accuracy level stayed same (73%) while loss level increase (58%).

  • Attempt #2 In this attempt, number of epoches was increased to 150 from 100.The new hidden layer that was added in attempt 1 stayed same. After change in the epoch numbers, the accuracy level stayed same (73%) while loss level increase (58%).

  • Attempt #3 Last attempt was made by adding two more dense layers to original model. This new model has 73% accuracy and 58% loss.

Summary:

None of three optimization attempts had acieved increase in accuracy, instead they were cause to increase in loss. Since there are a lot of categorical data in our dataset, may be trying decision tree or random forrest methods will work better.