Alphabet Soup Deep Learning Model for Funding Success Prediction

Overview of the Analysis

The purpose of this analysis is to create a binary classification model using deep learning techniques to predict if an organization funded by Alphabet Soup will be successful in their venture. The model utilizes a dataset of over 34,000 organizations that have received funding from Alphabet Soup, containing metadata about each organization.

Results

Data Preprocessing

• Target variable(s) for the model: The target variable for the model is IS_SUCCESSFUL.

• Feature variable(s) for the model: The feature variables for the model include APPLICATION_TYPE, AFFILIATION, CLASSIFICATION, USE_CASE, ORGANIZATION, STATUS, INCOME_AMT, SPECIAL_CONSIDERATIONS, and ASK_AMT.

• Variable(s) removed from the input data: The EIN and NAME columns were removed from the input data as they are identification columns and not useful as features or targets.

• Feature variable NAME has been brought back in the last model.

Compiling, Training, and Evaluating the Model

• Neurons, layers, and activation functions selected for the neural network model and rationale: The model consists of three hidden layers with 80, 30, and 1 neurons, respectively, and ReLU activation functions. The output layer uses a sigmoid activation function for binary classification. The structure was chosen to provide a balance between complexity and the potential for overfitting, while maintaining the ability to learn complex patterns in the data.

This model did not achive desired accuracy of 75%. The accuracy achived was 72.3%.

In this project, we ran about 4 models. The first 3 models removed the EIN and NAME columns and with applying difirrent neurons and layers and binning as bellow:

• Attempt1: Use same structure but different number of nurons in each layer and increasing the Epoch from 50 t0 100.
  • Accuracy increased very slightly to 72.7%
• Attempt 2: Optimising the structure using the Keras Tuner
  • Allow activation function to choose between relu, sigmoid, tanh
  • sigmoid is still the only option for the final layer
  • Allow number of neurons to vary from 6 to ~75
  • Accuracy increased to 73.1%
• Attempt 3: Try using few neurons (< number of features) with sigmoid activation function for non input layer
  • Accuracy decreased to 72.6%

The details of the 3 first attempts are provided in the AlphabetSoupCharity_Optimisation.ipynb.

• Attempt4: the final attempt that can be find in the Final-AlphabetSoupCharity_Optimisation.ipynb has the best accuracy of 78.5%.

Our Feature variable(s) for the model: The feature variables for the model include APPLICATION_TYPE, AFFILIATION, CLASSIFICATION, USE_CASE, ORGANIZATION, STATUS, INCOME_AMT, SPECIAL_CONSIDERATIONS, NAME and ASK_AMT. and Target variable(s) for the model: The target variable for the model is IS_SUCCESSFUL. The model consists of three hidden layers with 14, 7, and 1 neurons, respectively, The output layer uses a sigmoid activation function for binary classification and used Relu for other layers.

• Achievement of the target model performance: The model did achieve the target performance of 78.5% accuracy. However, multiple attempts were made to optimize the model, including adjusting input data, modifying the structure of the neural network, and modifying the training regimen.

• Steps taken in attempts to increase model performance: To increase model performance, the following steps were taken:

  • Dropping additional irrelevant columns from the input data.
  • Creating more bins for rare occurrences in columns and adjusting the number of values for each bin.
  • Adding more neurons to a hidden layer.
  • Adding or removing hidden layers.
  • Using different activation functions for the hidden layers.
  • Increasing or decreasing the number of epochs in the training regimen.

Summary

The deep learning model achieved the desired performance of 78.5% accuracy in predicting the success of Alphabet Soup-funded organizations. Several attempts were made to optimize the model through data preprocessing and neural network structure adjustments, ultimately leading to this improved performance.