mrvillafria / neural_network_charity_analysis

Creating a deep-learning neural network to analyze and classify the success of charitable donations

Neural Network Charity Analysis

Overview of Project

For this week's project, we will be implementing Neural Networks using the TensorFlow platform in Python. Neural Network is a machine learning technique that is modeled after the neurons in the brain. With Neural Networks, we can combine multiple statistical and machine learning models with little effort. This model helps evaluate all types of input data and produce a decision-making result.

Purpose

The purpose of this week's project is to design and train a Neural Network model so Alphabet Soup can predict which organizations are worth donating and which are too high risk. Alphabet Soup is a non-profit foundation dedicated to helping organizations that project the environment, improve people's well-being, and unify the world.

Results

The Alphabet Soup provided us a dataset containing more than 34,000 organizations that have received funding from Alphabet Soup over the years. Here are the columns that captured metadata about each organization:

• 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 consideration for application
• ASK_AMT—Funding amount requested
• IS_SUCCESSFUL—Was the money used effectively

Data Preprocessing

• The target variable for the data set was IS_SUCCESFUL.
• The variables we considered features were APPLICATION_TYPE, AFFILIATION, CLASSIFICATION, USE_CASE, ORGANIZATION, STATUS, INCOME_AMT, SPECIAL_CONSIDERATIONS, and ASK_AMT.
• The EIN and NAME columns were dropped since they were not target variables or features needed for the analysis.

Compiling, Training, and Evaluating the Model

In the image above, we can see the design of original Neural Network model. This consisted of:

• 2 hidden layers
  • First hidden layer
    • Neurons: 80
    • Activation function: ReLU
  • Second hidden layer
    • Neurons: 30
    • Activation function: ReLU
• 1 output layer
  • Output layer
    • Neurons: 1
    • Activation function: Sigmoid

It is a good rule of thumb for the initial model to use two to three times as many neurons as there are input features or values. We also selected slightly more complex activation functions for our hidden layers than our output layers for our initial model.

We were unable to achieve the 75% accuracy target model performance on our initial model.

A few steps we took to try to increase model performances were:

• Increasing the hidden layers from 2 to 3
• Changing the activation function of hidden layers or output layers to tanh
• Adjusting the neurons per hidden layer
• Dropping additional features that might have caused noise such as AFFILIATION_Other and USE_CASE_Other

Unfortunately after attempting to optimize our model 3 times, we were not able to increase the predictive accuracy.

Summary

After running several tests, we were unable to increase the model performance accuracy to 75%. A way to improve our accuracy could be to increase our dataset since our data may be insufficient or adjust the number of training epochs. Another model that we could potentially use would be Random Forest Classifier. Random Forest Classifiers are a type of ensemble learning model that combines smaller models into a more accurate model. The output may be similar to a Deep Learning model, but Random Forest Classifier can train and predict much faster.