This project was originally hosted on GitHub and can still be accessed there:
https://github.com/Ethiwam/iris-detect
This is a simple project meant to teach me how to build a neural network. I was in a
class on the application of AI and machine learning, when half way through the
semester we had to learn about building neural networks. Before starting this project
I thought, "wait, I also need to practice some PostgreSQL," and thus this project was
created.
I used the classic Iris dataset (Fisher, 1936), saving it all to a PostgreSQL database
and using SQL commands to interact with it. It's a small dataset, only with 150
samples, 4 features each, with 3 classes each sample could fall into. I designed a
deep neural network featuring 3 dense hidden layers, all of which used a sigmoid
activation function. The output layer uses a softmax function to output to one of
three classes: setosa, versicolor, or virginica. It's fairly straightforward, all
samples were normalized and the labels were all one-hot encoded before training the
model.
This does not save any keras models, though could easily be modified to do so. Every
time the code is run, a new model is trained on the training data, fine tuned with
the validation data, and tested on the test data, and will typically yield a model
with between 85%-100% accuracy. The training data, validation data, and test data
all come from the original dataset, roughly 80% training, 15% validation, 5% testing.
If you're interested in testing this code, below I've included both the neural network
model creator (the Python code) and a JSON file you can import into PostgreSQL that
contains the Iris dataset I used.
iris_classification_nn.py
iris_dataset_202404012309.json
Iris Detect
