Testing a new pattern recognition algorithm with a medical data-set

AiSara pilot

Zaim Awang
Jul 14, 2018 · 2 min read

I am trying out AiSara. Data: Heart disease data-set from approximately 300 patients from University of California Machine Learning Library. I am always most interested in practical application.

After minor cleanup for missing data, the data-set was uploaded into AiCentrica and dropped into AiSara. AiSara stands for Artificial Intelligence for Solution Approximation with Robust Algorithm. The idea here is to test the artificial intelligence algorithm with a real data-set coming from outside the development team. The time it took for me to upload the data-set and getting a full analysis is probably less than 2 minutes. The analysis only takes about 10 seconds top.

Annotated AiSara screenshot

Immediate findings:

In plain words, bear in mind I am not a medical professional, the chest pain (this is common sense) and the ST slope are most correlated to the heart disease. See ranking in the left panel in the image. AiSara default 3D visualization also automatically adapt by putting these 2 out of 3 variables in the x and y axis of the plot. The z axis is the heart disease diagnosis.

What is unclear to me is why a value of 3 (non-angina pain) and 4 (asymptotic pain ) has a more direct correlation to the diameter narrowing as oppose to typical angina pain. Maybe a medical doctor can explain or is it something to do with the data?

The data-set itself is not clear cut to me with even chest pain has an overall 15% impact only on the diagnosis (see left panel under the word Impact in the AiSara screenshot), and there maybe bias in the data with more data with chest pain values of 3 and 4.

Note: No conventional machine learning or neural network was harmed or utilized in AiSara algorithm.

Was this helpful? You can follow me on Twitter or check out our page (aicentrica.com)

Zaim Awang

Written by

Creator of Aisara | AiSara.ai | Algorithmist | Aspiring Author | Enterpreneur | Fluid Engineer | Family Man