perusmallit

Perusmallit, in Finnish research usage, refer to basic reference models used as benchmarks to compare more complex approaches. The term combines perus- meaning “basic” or “standard” and malli meaning “model.” As a category, perusmallit provide a transparent point of reference for evaluating predictive performance, interpretability, and data quality across disciplines from statistics and econometrics to psychology and machine learning.

Common forms include simple, well-known models such as the mean predictor, linear regression for numeric targets, logistic regression for binary outcomes, and Naive Bayes for text or categorical data. In time-series contexts, baselines may be an autoregressive model such as AR(1) or a seasonal naive forecast. The exact choice depends on data type and the research question.

Perusmallit are not substitutes for domain-specific or novel models. Their value lies in providing a baseline to quantify gains from added complexity and to aid interpretation. Proper evaluation is essential, typically involving cross-validation or holdout testing, and careful feature handling to prevent data leakage and biased comparisons.

In Finnish methodological writing, perusmallit are commonly discussed in methods sections to justify modeling decisions and to ensure comparability of results across studies. See also baseline models and benchmark approaches in data analysis.