Data Scientist how-to
How to Build and Evaluate a Prediction Model
Good modeling is about honest evaluation.
Quick answer
Good modeling is about honest evaluation. In short: define; split; train, then complete the remaining steps below.
Step by step
Define the question and success metric.
Split data into training and test sets.
Train a simple baseline model first.
Try more complex models.
Evaluate on the test set only at the end.
Explain results and limitations.
Safety reminders for data scientists
- Protect personal and sensitive data
- Follow privacy laws and company policies
- Check models for bias and fairness
- Take screen breaks and use ergonomic setups
Key terms
Dataset · Feature · Model · Training data · Overfitting · Regression: see the full data scientist glossary.