Getting Started With Orange 10: Feature Scoring and Ranking Video Lecture | Orange Data for Beginners - Data & Analytics

In a previous video on principle component analysis
we used wine data set.
These data are the result of a chemical analysis of wines,
grown in the same region in Italy,
but derived from three different cultivars, or vines.
The data include 13 features
reporting on quantities of chemical components.
How do I know which chemical components
are the most significant for differentiating between these vines?
With Rank of course!
This widget scores features with several scoring methods
based on the relation with class.
Connect File widget with Rank.
Rank displays two scoring methods as a default,
but we can display more if we want.
Say I want to see the scores for Gain Ratio,
Gini and ReliefF.
Now I want to select the features with the highest ReliefF score.
By default, the top five features are selected
and are already on the output.
Now I want to see how well are these features related to class.
Let's use some visualizations for that.
Connect the Box Plot to Rank
and inspect the first feature.
Use Group By 'Wine'
to see the results for each class separately.
Box Plot reports on mean, median,
variance and quartiles of each feature.
Mean is displayed as a vertical blue line.
Median is yellow.
The blue highlight denotes variance.
While the dotted lines display the first
and the fourth quartile.
The wines seem to have very distinct distributions of flavanoid concentration.
Seems like this feature separates the class very well.
But there must be an even better way of inspecting our features!
How about distributions? Let's see.
Distributions widget displays a value density plot for a given feature.
We can display value distributions for each class separately.
For flavanoids, these three distributions seem to be well separated.
Flavanoids are likely one of our most important features in the data set
as the separation is less pronounced with other features.
Rank widget can score and rank features
both for classification and regression.
Say we want to analyze a bit dated 'housing' data set,
where we would like to check which feature best correlates
with the house price in Boston suburbs.
Seems it's the economic status of inhabitants
and the average number of rooms.
Sort of obvious,
but it's still great to see this directly from the data.
Today we've learned how to determine
which features are the most interesting in our data set
and how to use feature scores
for plotting interesting visualizations.
Almost every data mining problem describes the data with features,
thus making feature scoring one of the best-loved techniques in the field.