9.10 Glossary of Terms
AIC (Akaike Information Criterion): A metric that balances model fit against complexity. Lower AIC values indicate a better trade-off between explanatory power and parsimony.
AutoML (Automated Machine Learning): Platforms that automatically test multiple model configurations — algorithms, hyperparameters, and feature sets — to identify the best-performing model.
BIC (Bayesian Information Criterion): Similar to AIC but with a stronger penalty for model complexity, favoring simpler models.
Cross-Validation: A technique for evaluating model performance by dividing data into k subsets, training on k-1, and testing on the remaining one, rotating through all subsets.
Descriptive Model: A model that summarizes and characterizes data to explain what has happened. Techniques include clustering, association analysis, and summary statistics.
Feature Engineering: The process of creating new input variables from existing data to improve model performance.
Feature Selection: The process of choosing which input variables to include in a model, balancing predictive power against complexity.
Generalizability: A model’s ability to perform well on new, unseen data — not just the data it was trained on.
Hyperparameter: A setting that controls the learning process of a model (e.g., learning rate, regularization strength) and is set before training, not learned from data.
Overfitting: When a model is too complex and learns noise in the training data rather than the underlying signal, resulting in poor performance on new data.
Parsimony: The principle of preferring the simplest model that adequately explains the data, reducing the risk of overfitting and improving interpretability.
Predictive Model: A model that forecasts future outcomes based on historical patterns. Techniques include regression, time series analysis, and machine learning.
Prescriptive Model: A model that recommends optimal actions by evaluating scenarios and constraints. Techniques include optimization and simulation.
R² (R-squared): The proportion of variance in the outcome variable explained by the model. Ranges from 0 to 1, with higher values indicating better fit.
RMSE (Root Mean Squared Error): A measure of the average magnitude of prediction errors, in the same units as the outcome variable. Lower is better.
Train/Test Split: Dividing a dataset into a training set (used to fit the model) and a test set (used to evaluate its performance on unseen data).
Underfitting: When a model is too simple to capture the real patterns in the data, resulting in poor performance on both training and new data.