Factor Analysis
Probabilistic linear model that separates signal from noise
Factor Analysis models the observed features as linear combinations of latent factors plus noise. Unlike PCA, it explicitly separates shared variance (factors) from unique variance (noise) per feature.
When to use:
- Psychometrics, social science, or survey data with latent constructs
- When modeling measurement noise separately from true signal is important
- Identifying underlying drivers shared across multiple observed variables
Input: Tabular data with the feature columns defined during training Output: Latent factor scores for each row
Model Settings (set during training, used at inference)
N Components (default: 2) Number of latent factors.
Max Iterations (default: 1000) Maximum EM iterations during training.
Tolerance (default: 0.01) Convergence threshold.
Inference Settings
No dedicated inference-time settings. The trained factor loadings project new data into the latent factor space.