Proper Feature Selection for Supervised Machine Learning

- Attributes:

- image:

+ ontology:
- description:

Proper Feature Selection for Supervised Machine Learning

Abstract:How can supervised machine learning algorithms be enhanced in their effectiveness to improve their predictive power by requiring proper feature selection; thus, advancing our understanding of still obscurely regulated complex phenomena, such as the gradual physiological decline due to aging?
Authors:Thomas Hahn [1], Dr. Daniel Wuttke [2], Philip Marseca [3], Dr. Richard Segall [4], Dr. Fusheng Tang [5]
[1]University of Arkansas at Little Rock, Little Rock, AR, USA
[2]Wuttke Technologies, Palo Alto, CA, USA
[3]SAP, Newtown Square, PA, USA
[4]Arkansas State University, Jonesboro, AR, USA
[5]University of Arkansas at Little Rock,, Little Rock, AR, USA

Our key observation for this discovery was that we humans are very bias pertaining to what kind of information (i.e. seeing, hearing, feeling, tasting and smelling) we pay attention but disregard the rest. That is why we don't get to see the whole picture and hence, remain stuck in a confusing state of mind - sometimes for many years - because we tend to neglect to consider almost all information outside the spectrum of our very subjective and limited range of sensory perception. Since we humans tend not to be systematic in selecting the data sources and dimensions of information (i.e. feature selection) without even being aware of it, we need the help of less bias artificial intelligence (AI).

The fact that we cannot perceive this information does not make it any less important or relevant for our lives. This was realized while trying to get a better understanding of the regulation of the aging process. The problem is that there are no omics datasets to test new aging hypotheses. This implies that nobody before us seemed to have felt that collecting transcriptome, proteome, metabolome and epigenetic data every 5 minutes throughout the entire lifespan of the yeast, would be worth the effort. We are totally capable of generating the data needed to advance in our understanding of aging and many other complex and still obscure phenomena, but the wet-lab scientists, who design our biological and medical experimental studies, don't seem to be even aware of missing something very important.

A good example is the magnetic field. We humans tend to ignore it because we cannot feel it. But, nevertheless, it affects our lives. It can cure depression. It can cause involuntary muscle twitching. Some birds use it to navigate the globe on their seasonal flight migration.

We are concerned about that there are other fundamental phenomena similar to the magnetic field, of which none of us is aware yet, because so far we have not tried to look for similar imperceptible information carrying dimensions.

For example, spiders, ants and bets are blind. However, visible light affects their lives regardless whether or not they have a concept of vision, since they have never experienced it. There could be other information carrying dimensions that – like the light for the spiders, ants and bets – is an imperatively hidden object (IHO), although it affects our lives so profoundly that we cannot understand aging and many other complex phenomena without considering such kind of information as well. That is why we recommend using artificial intelligence to reduce our observational bias.

Often, scientific progress has been made by accident. The means that a mistake, which changed the otherwise constant experimental environment in such a way that an unexpected result or observation was the consequence, was what helped us unexpectedly to finally make some progress. That is why we propose to intentionally vary external experimental conditions, methods, measurements, study designs, etc., to discover new features, which affect the outcome, much sooner.

The theories about the impact of still imperatively hidden objects (IHO) below might be challenging to understand, but it is worth trying, because if it succeeds, it will fundamentally improve our experimental methods and scientific study design.

- isDefinedBy:

- Relations:

- Add Property:

- Comments: