Concept - Linear Regression

Definition (s.249): hypothesis $h(x)=w_0+w_{1}x$ (univariate linear regression); trained from a training set by a learning algorithm. Also written $f(x)=w\ast x+b$ (s.258).

• Continuous target example: previous GPA -> score in H6751 (s.250).
• Binary target (pass=1/fail=0) makes it a classification problem (s.251).

Problems for classification (s.253-256)

• Output unbounded: with $y=0.3759x-0.59$, x=5 gives 1.289, x=1.2 gives -0.13 (s.255).
• Sensitive to imbalanced data: 20 extra points at GPA 6-7 shift the line and misclassify a GPA-3 passer (s.256).
• Remedy: pass the score through a sigmoid -> Concept - Logistic Regression.

Related

Concept - Logistic Regression, Concept - Sigmoid Function

Sources: 03 Text Classification - Chunk 09 (s.241-270)