In the dynamic world of forecasting, choosing the right variables, also called leading indicators or features, can make all the difference between accurate and unreliable forecasts. Our advanced variable selection tool empowers you to identify the most relevant variables, optimizing your forecasting models for accuracy, efficiency, and reliability.
Many organizations rely on correlation to identify leading indicators, but this approach often falls short in producing accurate forecasts.
Correlation only measures the linear relationship between two variables, whereas advanced methods can assess interactions among multiple variables, quantify the contribution of each, and account for group effects. This leads to significantly improved forecast accuracy.
For more insights, check out our interview with Professor Sune Karlsson, a key contributor to research on Bayesian Variable Selection.
In this recorded webinar, we will explore the advantages and disadvantages of various methodologies for identifying leading indicators. We'll cover approaches ranging from visual plotting and correlation analysis to advanced techniques for variable selection.
Variable selection is the process of choosing which variables (features) your model should actually use. Things like price, promotions, weather, holidays, macro indicators, or custom business signals. Instead of feeding the model every possible variable, we keep the signals that add predictive value and drop those that add noise.
Our feature offers several strategies to choose variables and transformations. It can use search algorithms (backward, forward, stepwise) to test many variable combinations, Lasso to shrink small coefficients to zero, and Bayesian methods that keep variables with high posterior inclusion probability.
Yes, you can override the variable selection results if you need to have specific variables in your forecasting models.
Multicollinearity mainly affects classical statistical models, while Lasso and Bayesian approaches already penalize it. For classical models, you can drop variables flagged in multicollinearity warnings or let variable selection remove them using a model that is sensitive to multicollinearity.
In Indicio, variable selection is applied only to multivariate models. Univariate models can only include other variables through exogenous modeling, which needs forecasts and would introduce look-ahead bias during evaluation since actual values are used for the exogenous variables.
Indicio offers several methods for ranking variables by relevance. It can either be done in the variable selection, where we use search algorithms (backward, forward, stepwise) that test variable combinations, Lasso to shrink small coefficients to zero, and Bayesian methods that keep variables with high posterior inclusion probability.
Ranking the variables' relevance can also be done in the last step in the forecasting process to translate complex forecast models into drivers and barriers using SHAP.
Indicio limits overfitting in several ways; train/validation splits and cross-validation, regularization (Lasso and Bayesian shrinkage), and automated variable selection that removes weak or redundant predictors.
Tip: comparing in-sample and out-of-sample results helps spot overfitting.
Yes. You can inspect diagnostics like coefficients, and impact on accuracy. Together these show which variables were kept or dropped, how strongly they influence the model, and whether they help or hurt forecast performance.
Variable selection adds some overhead, since it needs to test and compare different subsets of predictors. At scale, that cost is offset by smaller final models: fewer predictors speed up training of the chosen model and reduce inference latency in production.
Indicio automatically detects and treats missing values and seasonality. You can also flag and handle outliers and calendar effects such as holidays to further improve model performance.
.gif)
Experience the ease and accuracy of Indicio’s automated forecasting platform firsthand. Click to start a virtual demo today and discover how our cutting-edge tools can streamline your decision-making process.
We provide automated forecasting software that blends cutting-edge academic methods with market-specific leading indicators. This helps your organization reach the highest level of forecasting accuracy.
© 2025 Indicio Technologies All Rights Reserved