Lecture 30

• Lecture 30
  • Classification
  • Support Vector Machine
  • Decision Trees

Classification

• find linearly separable line (2D data me)
• line->plane->hyperplane
• curve->surface->hypersurfaces
• function fx nikalna bas mainly
• training phase
  • model built
• testing phase
  • kitna bdia model checked
• email spam
• confusion matrix
• rows: actually spam/non-spam
• cols: model says spam/non-spam

• how many times classified correctly = 72+20 = 92 = 92%

• top-right, spam, marked as non-spam

• bottom-left, non-spam (genuine mail) but marked as spam, hmm gandaa

• here, confusion matrix just store numbers, no severity/number

• so create a new matrix store that cost as well, which would affect final accuracy

Support Vector Machine

• is data linearly separable, say yes
• how many lines possible?
• infinite
• which one to pick

• pick best one.. waaah

• pick line maximizing margin

Decision Trees

• leaf nodes tell result
• baki nodes tell criteria/decision