Please enable JavaScript.
Coggle requires JavaScript to display documents.
Ch19: Interpret Model (feature impacts on target (overall impact of…
Ch19: Interpret Model
feature impacts on target
overall impact of feature without consideration of the impact of other features
treats each feature as standalone affect on target
focus attention on features most likely to yield additional predictive value
overall impact of a feature adjusted for impact of other features
calc of value of each feature in context - random shuffles
most important feature is scaled to 100%, others are scaled relative to it
directional impact
whether the presence of a value helps the model by assisting it in predicting readmissions or non-readmissions
partial impact
takes longer time
consider multiple models while calculating the values of the features therein
the power of language
diagnosis codes
filter stop word for text heavy alaysis
stop words have little value
hotspots
shows the most relevant (up to 4) combinations of features and their effect on the target
set of Venn diagrams where largest/most overlapping are organized in middle
hot/cold
hot = red (readmissions)
cold = blue (non readmissions)
reason codes
left and right thresholds allow for the specification of the probability cutoffs
computing reason codes is slower than computing predictions - reason codes engage in additional evaluations of why a prediction was set as the give probability for that case