Supervised Machine Learning

• Label: variable we are predicting

• Features: input variable describes our data

• Example: particular instance of data

• Labeled example: feature and label

• Unlabeld example: feature but no label

• Model: maps examples to predicted labels

Modeling Process

1. Training model on training set

   • Fit the model

2. Evaluate model on validation set

   • Estimate prediction error

3. Repeat

4. Record the result on test set

   • Assessment of the error

Metrics

• True positive: correct

• True negative: correct

• False positive: identify as positive but its not

• False negative: identify as negative but its not

• Accuracy: correct classfication / total classifcation

• Precision: TP/(TP+FP) correct/everything classfied as positive

• Recall: TP/(TP+FN) correct/all actual positives

• False positive rate: FP/(FP+TN)

• F1: 2precisionrecall/(precision+recall)

Loss function

• A loss function quantifies difference betwen actual response variable and the response variable predicted by your model