Demystifying Machine Learning

Summary:

• You need to compare your model’s performance against something else.
• Baseline models: A basic model upon which you want to improve.
  • Random predictor: Lower bound baseline
  • Most frequent class predictor: Always predict the most frequent class label
• Upper bound: The best possible performance your model can achieve
  • Usually indicate the human performance on the same task and dataset
• A confusion matrix is useful for analysing the output of a classifier by class