Please enable JavaScript.
Coggle requires JavaScript to display documents.
Ch. 3 Predictive Modeling (Models, Induction, and Prediction (Models…
Ch. 3 Predictive Modeling
Models, Induction, and Prediction
Models
Representation of reality
predictive model
formula for:
estimating the target
logical statement
mathematical
hybrid of logical and mathematical
judged on predictive abilities
descriptive modeling
gain insight into process or phenomenon
judged on intelligibility
Prediction
estimate unknown value
supervised learning
model creation
describes relationship between:
set of selected variables:
data example or instance
assume values present
has attributes or features
target
Model Induction
creation of models from data
procedure called:
learner
induction model
Induction algorithm
training data/labeled data
Supervised Segmentation
select informative values/attributes
pure or homogeneous w/respect to target
segments based on purity measure
common splitting criterion: info gain
based on
entropy
corresponds to impurity of set
high impurity = high entropy
=-p1
log(p1) - p2
log(P2)-
p = probability at each property
zero entropy = min disorder
one = max disorder
Measures disorder
IG measures change in entropy
how good is the variable?
Complications
attributes rarely split a group perfectly
not all attributes are binary
numeric attributes
rare to find pure data set
Classification Tree
composed of nodes
terminal node
creates leafs or segments
class prediction
interior nodes
predictive model
regression problems
problems with numeric values
use variance measure
Visualizing Segmentations
creating rules from trees