Joseph Azanza
Joseph Matthew R. Azanza
Data Scientist for a US-based cloud communications provider,
with expertise in Machine Learning, Artificial Intelligence,
Data Wrangling, Data Storytelling,
Sales Analytics, Sales Operations (Ops),
Marketing Analytics, Marketing Ops,
Business Intelligence, Strategic Initiatives,
Molecular Biology, and Biotechnology,
MS in Data Science
Asian Institute of Management
BS in Molecular Biology and Biotechnology
University of the Philippines Diliman
Machine Learning in Biology: Predicting Heart Failure based on Patient Medical History and Blood Chemistry
Joseph Azanza
Asian Institute of Management
Executive Summary
-
We can use the Leave One Out Cross Validation method to handle datasets with small samples as the calculated test accuracies are more robust than the train-test split method
-
Application and interpretation of ML in a biological problem is highly context dependent. In this study, while I was able to create an ML model with an 85% accuracy (trained on data from Pakistani patients), applying this model to predict heart failures for non-Pakistani people should be handled with skepticism. There are some context changes on a molecular level that even traditional machine learning models might not account for.
-
The value that the ML model with 85% accuracy gives when we applied in a different context is that it can provide us with a base expectation of what can happen. The ML model is used to augment decision making but not replace the decision makers.
Source code available at ML1_indiv