Page 55 - Reclaim YOUR DIGITAL GOLD (without audio)

Data ColleCtion Harvesting

Garbage Out,” which essentially means that if the data
provided is incorrect, the information generated as a
result of its analysis will be incorrect as well.
Let us use an example to demonstrate:

Here’s an interesting story about faulty data. A healthcare
project was underway with the goal of lowering the cost
of treating pneumonia patients. Based on their mortality
risk, automated machine learning (ML) was used to sort
through patient records to determine which patients
should receive antibiotics at home and which should
be admitted to the hospital. The ML was trained using
accurate historical data from many clinics, and the
resulting algorithm was dependable.

However, there was one significant exception to this
rule. Asthma, one of the most dangerous illnesses that
can accompany pneumonia, is almost always treated
by medical professionals in intensive care, resulting in
asthma patients having a significantly lower risk of death.
As a result, the algorithm concluded that asthma is not
as dangerous during pneumonia because there were no
fatal asthmatic cases in the data. Consequently, despite
having the highest risk of pneumonia complications, the
algorithm recommended that asthmatics be sent home.

Data is essential for machine learning. It is the single
most important factor that allows algorithms to be
trained and explains why machine learning has grown
in popularity in recent years. However, regardless of the
actual terabytes of data available or the individual’s skill
level in the field of data science, if the individual is unable
to make sense of the data records, a machine is nearly
worthless, and it may even be dangerous.

35