Machine learning mistakes

Machine Learning for Business

Karolis Urbonas

Head of Machine Learning & Science, Amazon

Mistakes

  • Machine learning first
  • Not enough data
  • Target variable definition
  • Late testing, no impact
  • Feature selection
Machine Learning for Business

Machine learning first

pyramid-raw

Machine Learning for Business

Not enough data

Preparation

Machine Learning for Business

Target variable definition

  • What are we predicting?
  • Can we observe it?
    • Contractual churn - customer terminated the premium credit card
    • Non-contractual churn - customer started using another grocery store
  • In-depth analysis
  • Business field expertise
Machine Learning for Business

Feature selection

Inference (what affects the target variable?)

  • Choose variables that you can control (website latency, price, delivery, customer service etc.)
  • Business has to be involved in feature selection

$$

Prediction (can we estimate the target variable value in the future?)

  • Start with readily available data
  • If model performance is OK, test it
  • Introduce new features iteratively
Machine Learning for Business

Late testing, no impact

abtest

Machine Learning for Business

Let's practice!

Machine Learning for Business

Preparing Video For Download...