Neural Network Model

• 

  Model

Inspiration

With an emerging interest for Machine Learning and neural networks in Healthcare and Medicine I wanted to build something meaningful and applicable to the real world. I wanted to work with one of the popular UC Irvine Machine Learning Repository data sets.

What it does

The model essentially takes the Breast Cancer data-set and performs some basic mathematical operations to predict if the tumours are malignant or benign. I used the two-class logistics regression algorithm as the activation function for the hidden layer. After the operation, I implemented scoring and evaluation of the model as well. The model has an approximate accuracy of 96%.

How I built it

I built the model using Microsoft Azure Machine Learning Studio. I had the opportunity to utilize the statistical functions, Machine Learning (Initialization, scoring, and evaluation), data transformation and data set features. I approached the development of the model in two large sections:

Part 1: Data Clean Up

• Imported data set from UC Irvine Machine Learning Repository onto Azure
• Modified the data set (removed certain elements like the ID row from CSV)
• Utilized math operations from statistical functions to remove rows with missing / invalid data

Part 2: Training

• Normalized the data-set (Z-score) and split into a training set and data set
• Initialized the model using the two-class logistics regression to deal with the issue of classification (malignant vs benign)
• Scored and evaluated the model in comparison to data set

Challenges I ran into

I had difficulty figuring out which algorithm would work best as an activation function for training the data set. Additionally, splitting the data set and developing a work around was a little tricky in Azure.

Accomplishments that I'm proud of

I am proud of finally gaining a better understanding of neural networks and their overall development. I am also quite pleased to have had my network model fully run / compile on Azure and publish it in my gallery.

What I learned

I learned to use Microsoft Azure and the plethora of features it includes that are very useful for modelling complex problems. I found it was quite easy to work with especially when there are difficulties in code implementation. Microsoft Azure made it easy to get my idea done with limited knowledge of writing out complex algorithms.

What's next for Neural Network: Breast Cancer Model

I would like to write out the results obtained from the model into a CSV file. Essentially, extending the final results in a more viewable format, that is easier to work with and analyze.

Built With

• azure
• machine-learning
• neural-network