scientific concept · BEGINNER LESSON
Prediction
Prediction is the act of using a model, algorithm, or mechanism to estimate a future observation or output from current information. In science, it also serves as a test of whether a model matches reality, often by checking accuracy on held-out data.
What prediction is
Prediction is an estimate of what will happen next, or what output a system will produce, based on current information. In data science, the usual target is a measurable outcome, and predictive accuracy is judged by comparing predicted values with testing data that were not used to fit the model. In mechanistic research, prediction can also mean using known parts of a system and their interactions to infer how the system will behave.
[2] The Scope Of Neural... [3] Veridical data science - PMC - NIH
How prediction works
Prediction can come from two different strategies. One strategy learns statistical relationships in data and checks them on held-out data. Another strategy uses mechanistic explanation: it identifies physical parts, the rules for their interaction, and the way those parts produce an output. The second approach aims to predict only from parts that are actually relevant to the mechanism, not from accidental correlations.
[2] The Scope Of Neural... [3] Veridical data science - PMC - NIH
Why more data does not always help
The Nature Physics source warns that data alone can fail in high-dimensional systems, where the number of variables is large. It notes that recurrence ideas from dynamical systems theory imply some predictability, but the recurrence time can grow astronomically with system dimension and required precision. In such cases, better prediction often needs theory, simplification, or conceptual insight, not just more data.
Limitations and interpretation
Prediction is useful as a reality check, but it is not the same as understanding. A model may predict well for the wrong reason, such as using a correlated variable that is not part of the mechanism. Also, prediction depends on choices about the model form, training data, testing data, and what counts as accurate enough. The sources emphasize that good forecasting can require giving up detail and choosing simpler models.
[2] The Scope Of Neural... [3] Veridical data science - PMC - NIH [1] The limits of machine prediction - Nature Physics
SEE THE IDEA
Two routes to prediction
Measurements used to train or test a model.
Learns patterns and is checked on held-out data.
Uses parts and interactions that produce the output.
The estimated future observation or system response.
Observed data can feed either a statistical model or a mechanistic model, and both aim to produce a predicted outcome that can be compared with new data.
Prediction can be built from data patterns or from mechanisms, but both must be tested against observations.
TRY IT YOURSELF
Test your understanding
Choose an answer, then check the explanation.
PUT IT TOGETHER
Match each concept to its meaning
Drag a label onto its description, or select a label and then select a description. This also works with a keyboard.
4 pairs to match
Sources and further reading
This general lesson was generated from public source material. It explains a concept; it does not interpret a private paper or provide medical advice.