yempi / neural-network-charity-analysis

neural-network-charity-analysis

Overview

In this project neural networks will be used to help predicting where to invest as a foundation. With the help of Machine Learning and Neural Networks we will create a binary classifier that is capable of predicting whether applicants will be successful if funded by the foundation (Alphabet Soup).

The data will be taken from a .csv file containing more than 34k organizations that have received funding in the past.

The process of completing this project is as follows:

• Preprocessing Data for a Neural Network Model
• Compile, Train, and Evaluate the Model
• Optimizing the Model

Results

• Data Preprocessing

  • Variables considered as targets: Column IS_SUCCESSFUL contains binary data pertaining whether or not the charity donation was used effectively.
  • Variables considered as features: Columns APPLICATION_TYPE, AFFILIATION, CLASSIFICATION, USE_CASE, ORGANIZATION, STATUS, INCOME_AMT, SPECIAL_CONSIDERATIONS, ASK_AMT.
  • Variables that should be removed: Columns EIN and NAME are considered identification information.

• Compiling, Training, and Evaluating the Model

  • Number of Neurons, Layers and Activation Functions selected for the model:

    Hidden Layers:

    • #1: 80 neurons.
    • #2: 30 neurons.

    Input Data:

    • 43 features
    • 25k samples

    Output Layer:

    • 1 neuron.

    Activation Functions:

    • ReLU: for speeding up the training process (hidden layers).
    • Sigmoid: considering a binary classification (output layer).
    • adam: optimizer for compilation.
    • binary_crossentropy: loss function for compilation.

• Target Model Performance

The goal was to meet or excede the 75% model accuracy, but this was not achieved obtaining numbers as close as 72%.

• Steps taken to increase model performance

1. Bucketing to the feature ASK_AMT and organizing the values by intervals.
2. Increasing the number of neurons in a single hidden layer, then using a model containing 3 hidden layers.
3. Trying the tanh activation function.

Summary

It was not a satisfactory result taking into account that 75% is a standard to measure the accuracy of a model and it was not achieved by this example. In the future a Random Forest Classifier could be tested to try and achieve better results considering that it is a binary classification output.