BCDxML

Initial Principal Component Analysis
Feature Extraction and Value Distribution
Feature Correlation Matrix
Two Branch Decision Tree for Random Forest Algorithm
Final Results

Inspiration

Having both come from families with histories of cancer, we understand how important early detection is for reducing mortality rate. Scheduling with medical professionals can cause delays in diagnosis (especially in third world countries where our families originally came from), which can only lead to over-analysis and worst of all -- fear. Hence, we created a rather accurate detection algorithm for one of the most well documented type of cancer out there -- breast cancer -- so that patients could have a good idea of what they're up against in advance.

What it does

Our Breast Cancer Detection algorithm through Machine Learning (BCDxML) takes data that the patient knows and feeds back a prediction of whether or not they have breast cancer based on their symptoms, along with the percent confidence in the diagnosis.

How we built it

We used pulled data from a publicly available data set distributed by the University of California, Irvine. From there, we used a principal component analysis to detect the most valuable variables within the data set, and from there used a random forest algorithm to build a model to diagnose the cancer.

Challenges we ran into

Lots of dependency issues, and performance issues training the model, however, these are all one-time concerns.

Accomplishments that we're proud of

over 97% accuracy in detecting the cancer correctly.

What we learned

A whole lot of R, and the basics of machine learning for branch prediction. The power of data, and the limitless future of machine learning for medical applications.