Please enable JavaScript.
Coggle requires JavaScript to display documents.
Chapters 17 & 18 (ch 17 Introduction (Must understand each algorithm…
Chapters 17 & 18
ch 17 Introduction
Must understand each algorithm
Performance characteristics more important than how it works
FVE Binomial explains = R^2
Receiver Operating Characteristics (ROC) Curve
Can view cross-validation and validation results
Two "mountains" created
One for true and one for false
Can show by density and frequency
Changes the shape of mountains
Called "confusion matrix"
Want mountains to be separate
Can change threshold after predicting
Generally for binary problems
True positive/negative = correctly predicted positive/negative
False positive/negative = wrongly predicted positive/negative
True Negative/ Positive Rate = probability of being correct given that "a" or "b" occurred
Negative predictive value = proportion of predicted negatives actually negatives
True Negative Rate = proportion of negative cases model capable of finding
False Positive Rate = probability of false positive
ROC Curve
Good ROC curves calculated by AUC
AUC = Area under the curve
Calculated using 100 small boxes and counting how many fit under curve
Sample algorithm Model
3 Steps
Find most predictive feature
split feature into 2 branches
Repeat step 2 for each new branch
Using the lift chart for business decisions
lift chart sorts all validation cases by probability of readmission
Want blue and orange lines that fully overlap
Sort data and create profit chart
Look for the cutoff point
Model Comparison
Use to see differences between models
Check ROC curves against each other
Perfect = straight line along top
Look at dual lift
Calculate by taking left model's prob prediction for case- right model prediction for a cals
Prioritizing model criteria and selecting model
Consider 5 criteria
Predictive accuracy.
How correct the model is
Prediction speed. How long it takes to make a prediction
Speed to build model. How long the model takes to be made
Familiarity with model. How much of an expert the data scientist is on the model
Insights. what conclusions you can draw from the model