- Create a deep-learning neural network to analyze and classify the success of charitable donations in order to determine the future decisions of the company—only those projects likely to be a success will receive any future funding.
A csv file containing more than 34,000 organizations that have received funding over the years.
Columns metadata:
-
EIN and NAME—Identification columns (irrelevant variables)
-
APPLICATION_TYPE—Charity Foundation application type (irrelevant variables)
-
AFFILIATION—Affiliated sector of industry (Features)
-
CLASSIFICATION—Government organization classification (Features)
-
USE_CASE—Use case for funding (Features)
-
ORGANIZATION—Organization type (Features)
-
STATUS—Active status (irrelevant variables)
-
INCOME_AMT—Income classification (Features)
-
SPECIAL_CONSIDERATIONS—Special consideration for application (irrelevant variables)
-
ASK_AMT—Funding amount requested (Features)
-
IS_SUCCESSFUL—Was the money used effectively (Target)
Total params: 8,903 Trainable params: 8,903 Non-trainable params: 0
-
In this deep-learning neural ntework model, there are total three hidden layers with 84, 50 and 20 neurons respectively. The input shape was (42,) and output layer's neurons was 1 (binary_classification). The optimizer is adam and loss metrics setting is binary_crossentropy.
-
The model's predictive accuracy was under 75%, which means the model was not able to predict correctly whether or not a target company will be re-funded over 57% of the time.
-
In order to try and improve the model predictive performance, I tried to remove outliers and noisy in ASK_AMT variable by using 3 times of zscore method. In addition, removing irrelerant variables provided a few improvement. remove_outliers.PNG
Despite of reducing dimensionality with PCA (Principal Component Analysis), the model's accuracy didn't significantly improve.
- I tried to add a LeakyReLU layer to deep-learning model, but it didn't work as I expected. Furthermore, adding some regularizers on nauron's kernel still failed.
The model's accuracy recording:
