Prediction vs. inference dilemma

Machine Learning for Business

Karolis Urbonas

Head of Machine Learning & Science, AWS

Inference vs. prediction dilemma

Inference or causal models:

  • The goal is to understand the drivers of a business outcome
  • Inference focused models are interpretable
  • Less accurate than prediction models

Prediction:

  • The prediction itself is the main goal
  • Are not easily interpretable i.e. work like "black-boxes"
  • Much more accurate than inference models
Machine Learning for Business

Start with the business question

  • "What are the main drivers of fraud?"
    • Inference
  • "How much conditions X impact heart attack risk?"
    • Inference
  • "Which transactions are likely fraudulent?"
    • Prediction
  • "Is the patient at risk of having a heart attack?"
    • Prediction
Machine Learning for Business

Modeling data structure

inference-vs-prediction1

Machine Learning for Business

Target variable

inference-vs-prediction2

Machine Learning for Business

Input features

inference-vs-prediction

Machine Learning for Business

Using input features

inference-vs-prediction4

Machine Learning for Business

Predicting target variable

inference-vs-prediction5

Machine Learning for Business

Inference model focus

inference-vs-prediction6

Machine Learning for Business

Prediction model focus

inference-vs-prediction7

Machine Learning for Business

Let's practice!

Machine Learning for Business

Preparing Video For Download...