Explain relationships with interpretable coefficients and rigorous diagnostics for trustworthy predictions
These outputs come directly from the analysis module.
R² (model fit 0-1), Adjusted R² (complexity-adjusted), RMSE & MAE (prediction accuracy in target units), AIC & BIC (model comparison), F-statistic (overall significance)
Impact of each predictor with 95% confidence intervals, p-values for significance testing, VIF for multicollinearity check, complete ANOVA variance breakdown
Residual plots for assumption checking, Q-Q plots for normality, Cook's distance for outlier detection, prediction intervals with uncertainty bands, fitted vs actual comparisons
Steps performed by the analysis module.
Builds a linear model from your target and features (formula interface).
Calculates R², Adjusted R², RMSE, MAE, AIC, BIC, F‑stat and p‑value.
Returns estimates with standard errors, t‑stats, p‑values, and 95% CIs.
Provides residual data, Q‑Q points, histogram bins, Cook’s D, and VIF (when available).
Coefficients with confidence intervals (Estimate, SE, t, p, 95% CI).
Residual plots (residuals, Q‑Q, histogram) and influence (Cook's D).
VIF for multicollinearity assessment (when available).
Tabular data frame for analysis.
Name of numeric target column (regression).
Vector/list of predictor column names.
User context and processing ID (for traceability).
Provide a dataset with numeric target and features (numeric or encoded categorical). Factors are supported; please provide encoded categoricals if needed. We review assumptions visually and return interpretable results.
Outputs include coefficients, intervals, metrics (RMSE/MAE/R²), and diagnostic plots for confidence in decisions.
From specification to diagnostics
Define the model, validate inputs, and review univariate/multivariate profiles.
Estimate OLS; produce residual, influence, and VIF diagnostics; adjust as needed.
Present coefficients and intervals; provide guidance on drivers, limitations, and next steps.
Linear regression provides fast, interpretable baselines for forecasting and driver analysis—great for stakeholder alignment.
Use OLS to quantify relationships and set expectations. If diagnostics show issues, we propose robust or regularized alternatives for stability.
Note: For nonlinearities/interactions, engineer features or compare with tree‑based models; for collinearity, consider regularization.
Get interpretable coefficients with solid diagnostics
Read the article: Linear Regression