Fundamentals of Data Analytics (32130) Week 8

Classifier

its job is to tell you the probability of how accurate it is.

classifier will create a line to separate all the values, instead of scatter plot, transfer the representation of indiv value to distribution representation = ROC curve

Confusion Matrix

Type I error FP

falsely predicted as positive, obvious errors

Type II error FN

was actually positive but were predicted negative

Measures of performance

Accuracy rate

the ratio of correct predicitions to all predictions.

Number of true positives and true negatives / total number of predictions

Error rate

the ratio of incorrect predictions to all predictions, (1 - accuracy). Number of false positives and false negatives divided by the total number of predictions.

ROC curve

Receiver Operating Characteristic (ROC) curve

uses accuracy and error rate

True Positive Rate vs False Positive Rate

straight diagonal line: the results are the same, can’t distinguish between positive and negative = Random Classifier

50% chance of getting the right answer - very bad model.

AUC

Area under the ROC curve

probability that the model will rank the positive higher than the negative

the accuracy

KNN

doesn’t need to specify learner and predictor node

The only one