A probability distribution describes how values, frequencies, or probabilities are spread across the possible values of a variable. In machine learning work, it helps determine whether data is plausible and whether a model can produce the required kinds and proportions of outputs. section 6.1.1 section 7.3.3
Use distributions as checkpoints during exploration and model selection:
- Check whether typical and extreme entities make sense, and investigate unexplained gaps or dead spots in value ranges. section 6.1.1
- Compare unusual periods, such as weekends or holidays, with stakeholder expectations before treating them as errors. section 6.1.1
- Confirm that the selected algorithm can represent the required output range and produce the correct output proportions in principle. section 7.3.3
From customer data to model choice
Suppose customer records show hundreds of weekly transactions for some entities and almost none for others. Inspect whether those extremes and the overall pattern fit the business context. If a model must produce a graduated control signal from 0.0 to 1.0, reject an approach limited to binary on/off output. section 6.1.1 section 7.3.3
